ELECTRA_Semantic_Text_Similarity.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "ELECTRA  Semantic Text Similarity.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "PADp4hO9_uhO"
      ],
      "toc_visible": true,
      "authorship_tag": "ABX9TyMxTY3sU/FavUVw03Jj98CJ",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU",
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "5b16ffff66c14c94972d877381c3616e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_66fef79997c24c70bbf172405b437210",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_f4e474becb154483ab4b127ffa98af0b",
              "IPY_MODEL_1c60f452c0e643d792b0b86e4dfe61e2"
            ]
          }
        },
        "66fef79997c24c70bbf172405b437210": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "f4e474becb154483ab4b127ffa98af0b": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_0f197cf5a5694dd4858cf6925acba372",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 231508,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 231508,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_42bba2c009e94aa2851c2f5b02467b0b"
          }
        },
        "1c60f452c0e643d792b0b86e4dfe61e2": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_5230bca55323461b8f7f6446c3b8dc17",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 232k/232k [00:00&lt;00:00, 573kB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_8745a7d59ac94df9bbdf297ab13733ec"
          }
        },
        "0f197cf5a5694dd4858cf6925acba372": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "42bba2c009e94aa2851c2f5b02467b0b": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "5230bca55323461b8f7f6446c3b8dc17": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "8745a7d59ac94df9bbdf297ab13733ec": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "ce3983cca06142d49cbbf101b28cf3d9": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_678f946a99af41968a18db63a841e0e9",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_6edf8b8caef149ab98271915a5126eac",
              "IPY_MODEL_b75609f89e264b059eb0a5f9a46cc7f6"
            ]
          }
        },
        "678f946a99af41968a18db63a841e0e9": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "6edf8b8caef149ab98271915a5126eac": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_fc4c682557204486b130ab67985ef32c",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 467,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 467,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_b06674f688a4438e8f7d11dd137ffe94"
          }
        },
        "b75609f89e264b059eb0a5f9a46cc7f6": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_950bb8f2c0794397a5125b3819e4a43d",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 467/467 [00:08&lt;00:00, 55.2B/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_c2228155806d4bf6b8f59d2e5f251e91"
          }
        },
        "fc4c682557204486b130ab67985ef32c": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "b06674f688a4438e8f7d11dd137ffe94": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "950bb8f2c0794397a5125b3819e4a43d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "c2228155806d4bf6b8f59d2e5f251e91": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "3957196b6c8b4572935d00dd78b3b056": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_8055d0ac75654a5ea8ebc1e712cf0ef1",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_3036b3bd4a4d4d1891530f8df259317e",
              "IPY_MODEL_863b14e4780c4452bd911dfcddfe4219"
            ]
          }
        },
        "8055d0ac75654a5ea8ebc1e712cf0ef1": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "3036b3bd4a4d4d1891530f8df259317e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_3acd58850e954797b9f87811504096e5",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 440343552,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 440343552,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_36591ad71cec45138c172aadfbd5d157"
          }
        },
        "863b14e4780c4452bd911dfcddfe4219": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_2261e2cd10f74c75ae19890db5a9b1d2",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 440M/440M [00:07&lt;00:00, 58.5MB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_6669284c25a64c79abfc63fd0b7377f2"
          }
        },
        "3acd58850e954797b9f87811504096e5": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "36591ad71cec45138c172aadfbd5d157": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "2261e2cd10f74c75ae19890db5a9b1d2": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "6669284c25a64c79abfc63fd0b7377f2": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "16228d22f42f41c7baae93d2f4f9faa8": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_5119b17f80854ffebe0e9458eacb778f",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_aaac04918b01412ab52a647df2ead4d3",
              "IPY_MODEL_2a2abc8d31a64c7e840b0ab9639c7136"
            ]
          }
        },
        "5119b17f80854ffebe0e9458eacb778f": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "aaac04918b01412ab52a647df2ead4d3": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_08b852d8acc148dcab7586104de73b56",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 467,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 467,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_1d0528b5c8ac41e7abd78e6ba51c36b1"
          }
        },
        "2a2abc8d31a64c7e840b0ab9639c7136": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_174bf646c7ec42caaace3472efd0cce9",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 467/467 [00:08&lt;00:00, 57.2B/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_8abcd4a7d6b34a38b13678538d945752"
          }
        },
        "08b852d8acc148dcab7586104de73b56": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "1d0528b5c8ac41e7abd78e6ba51c36b1": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "174bf646c7ec42caaace3472efd0cce9": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "8abcd4a7d6b34a38b13678538d945752": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "a56d60dae6544f75966da9c88832a432": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_e25554e1d03e44819c0a592cc76a9a40",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_514a29861a164e05b767d6b10bfc98b5",
              "IPY_MODEL_8f2efda139d74bdfbcefbe7dc9a61d6c"
            ]
          }
        },
        "e25554e1d03e44819c0a592cc76a9a40": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "514a29861a164e05b767d6b10bfc98b5": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_d8ca9f13f86d4957811b157af10551fc",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 440343552,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 440343552,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_6f294a984012499fa052e068531b51fa"
          }
        },
        "8f2efda139d74bdfbcefbe7dc9a61d6c": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_283a878e291a4cb99e4b4d5fe17d4967",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 440M/440M [00:08&lt;00:00, 54.9MB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_046a49a886044c02935ad4d5886cbeed"
          }
        },
        "d8ca9f13f86d4957811b157af10551fc": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "6f294a984012499fa052e068531b51fa": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "283a878e291a4cb99e4b4d5fe17d4967": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "046a49a886044c02935ad4d5886cbeed": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "c815b5730056477b83b7d8ddb23eebac": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_01d7c17f88b54ff2b0c7babef44191b8",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_e52809531432458da1ffdf92604dea38",
              "IPY_MODEL_a088d70f081d435ebf6dce0b45165df3"
            ]
          }
        },
        "01d7c17f88b54ff2b0c7babef44191b8": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "e52809531432458da1ffdf92604dea38": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_97f4ef1eb5034d1fbe2f8e8a9de0bdb2",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 433,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 433,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_861a215eb8d443daa8ca35fa8f57dbf2"
          }
        },
        "a088d70f081d435ebf6dce0b45165df3": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_8bfdd0e87d864af38e0dc64756304bb1",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 433/433 [00:06&lt;00:00, 65.9B/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_1bf583ee208547ed8e1f126863f47dba"
          }
        },
        "97f4ef1eb5034d1fbe2f8e8a9de0bdb2": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "861a215eb8d443daa8ca35fa8f57dbf2": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "8bfdd0e87d864af38e0dc64756304bb1": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "1bf583ee208547ed8e1f126863f47dba": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "ecc2382a18cb4b74b2c4f7823383637f": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_1ef3ada8af3d4f5fa058d0eb7071fe22",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_dab0d86719e843c68a947b127a470fd6",
              "IPY_MODEL_28f7db0ff5904863a3c94b95368ccc6a"
            ]
          }
        },
        "1ef3ada8af3d4f5fa058d0eb7071fe22": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "dab0d86719e843c68a947b127a470fd6": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_00f3e40d82bc4c128a6254636cdc7548",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 440473133,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 440473133,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_c96e6b0c61684aff967dc974143ab8c8"
          }
        },
        "28f7db0ff5904863a3c94b95368ccc6a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_14b14b227d8b4f0294c73e7527be743f",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 440M/440M [00:06&lt;00:00, 68.4MB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_8aee81ec8bb145a8b6f126b91d78e317"
          }
        },
        "00f3e40d82bc4c128a6254636cdc7548": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "c96e6b0c61684aff967dc974143ab8c8": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "14b14b227d8b4f0294c73e7527be743f": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "8aee81ec8bb145a8b6f126b91d78e317": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        }
      }
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/glingden/Natural-Language-Processing-NLP/blob/master/ELECTRA_Semantic_Text_Similarity.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "szrWqOlB9-R7",
        "colab_type": "text"
      },
      "source": [
        "# Electra Semantic Text Similarity\n",
        "In this notebook, the Electra (electra-base-discriminator) model is used for finding  semantic text similarity between the pair of sentences.  The [STS_B benchmark](https://ixa2.si.ehu.es/stswiki/index.php/STSbenchmark) dataset is used for this work. This dataset consists of separate training, validation and test data. The work in this notebook contains two parts:\n",
        "1. Sentence level embedding extraction from the pre-trained  bert model and apply cosine similarity between the pair sentences (No fine-tunned, but only using sentence vector representation from the pretrained model)\n",
        "\n",
        "2. Fine-tune: The STS_B train dataset is used for fine tunning the model and validaiton dataset is used for model validation. And, the fine-tunned model is used to predict the test dataset."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1xo41bKl99NK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#import necessary libraries\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import torch\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PFT-i2ACvRGA",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        },
        "outputId": "4163c898-a0e1-42ea-98d6-2fd38d4d638d"
      },
      "source": [
        "#check version\n",
        "print(torch.__version__)\n",
        "print()\n",
        "print(np.__version__, pd.__version__)\n",
        "!python -V"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1.5.1+cu101\n",
            "\n",
            "1.18.5 1.0.5\n",
            "Python 3.6.9\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6_AtOWHBi-Rv",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "7c207b47-2481-470a-ba39-a44cd0a786e0"
      },
      "source": [
        "#check GPU is available or not \n",
        "if torch.cuda.is_available():\n",
        "\n",
        "  device = torch.device(\"cuda\")\n",
        "  print('Found GPU :', torch.cuda.get_device_name(0))\n",
        "\n",
        "else:\n",
        "    print('Not found, use CPU instead')\n",
        "    device = torch.device(\"cpu\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Found GPU : Tesla T4\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bRXVmlCHi-Wn",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 122
        },
        "outputId": "aaa493a3-8af5-4ef8-e054-acc147567028"
      },
      "source": [
        "#mount drive \n",
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Go to this URL in a browser: https://accounts.google.com/o/oauth2/auth?client_id=947318989803-6bn6qk8qdgf4n4g3pfee6491hc0brc4i.apps.googleusercontent.com&redirect_uri=urn%3aietf%3awg%3aoauth%3a2.0%3aoob&response_type=code&scope=email%20https%3a%2f%2fwww.googleapis.com%2fauth%2fdocs.test%20https%3a%2f%2fwww.googleapis.com%2fauth%2fdrive%20https%3a%2f%2fwww.googleapis.com%2fauth%2fdrive.photos.readonly%20https%3a%2f%2fwww.googleapis.com%2fauth%2fpeopleapi.readonly\n",
            "\n",
            "Enter your authorization code:\n",
            "··········\n",
            "Mounted at /content/drive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4T322tS-jLB9",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 374
        },
        "outputId": "4698dbc5-a51e-488f-bef2-251fe73fa42a"
      },
      "source": [
        "#load dataset\n",
        "dataset_types = [ \"sts-train.csv\", \"sts-dev.csv\", \"sts-test.csv\",] # 3 datasets\n",
        "col_names = [\"genre\", \"file\", \"years\", \"_\", \"score(0-5)\", \"sentence_1\", \"sentence_2\"] #columns names\n",
        "\n",
        "#collect as a list of pandas dataframes\n",
        "df_list = []\n",
        "for dataset in dataset_types:\n",
        "  df = pd.read_csv(\"/content/drive/My Drive/Google_Colab/stsbenchmark_dataset/\"+ dataset, \n",
        "                 delimiter=',' , \n",
        "                 header= None,\n",
        "                 names= col_names\n",
        "                )\n",
        "  \n",
        "\n",
        "  df_list.append(df)\n",
        "\n",
        "#separate dataset (train, dev and test)\n",
        "df_train = df_list[0]\n",
        "df_dev = df_list[1]\n",
        "df_test = df_list[2]\n",
        "\n",
        "#show random 5 rows in each dataset (train, dev and test)\n",
        "print(\"Show train_data shape: {}\".format(df_train.shape))\n",
        "print(\"Show dev_data shape: {}\".format(df_dev.shape))\n",
        "print(\"Show test_data shape: {}\".format(df_test.shape))\n",
        "print(\"Show Random 5 rows:\")\n",
        "df_train.sample(5)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Show train_data shape: (5749, 7)\n",
            "Show dev_data shape: (1500, 7)\n",
            "Show test_data shape: (1379, 7)\n",
            "Show Random 5 rows:\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>genre</th>\n",
              "      <th>file</th>\n",
              "      <th>years</th>\n",
              "      <th>_</th>\n",
              "      <th>score(0-5)</th>\n",
              "      <th>sentence_1</th>\n",
              "      <th>sentence_2</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>2340</th>\n",
              "      <td>main-forum</td>\n",
              "      <td>deft-forum</td>\n",
              "      <td>2014</td>\n",
              "      <td>340</td>\n",
              "      <td>4.4</td>\n",
              "      <td>Breathe In, Breathe Out, Move On.</td>\n",
              "      <td>buwhahahaha Breathe In, Breathe Out, Move On.</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4445</th>\n",
              "      <td>main-news</td>\n",
              "      <td>headlines</td>\n",
              "      <td>2014</td>\n",
              "      <td>124</td>\n",
              "      <td>3.6</td>\n",
              "      <td>Bangladesh building collapse death toll hits 362</td>\n",
              "      <td>Bangladesh building-collapse death toll climbs...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4300</th>\n",
              "      <td>main-news</td>\n",
              "      <td>headlines</td>\n",
              "      <td>2013</td>\n",
              "      <td>687</td>\n",
              "      <td>3.2</td>\n",
              "      <td>Navy jet crashes into Virginia buildings, inju...</td>\n",
              "      <td>Navy jet crashes into Virginia apartments,9 hurt</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3474</th>\n",
              "      <td>main-news</td>\n",
              "      <td>deft-news</td>\n",
              "      <td>2014</td>\n",
              "      <td>24</td>\n",
              "      <td>3.2</td>\n",
              "      <td>russian officials have called for a conference...</td>\n",
              "      <td>antonov spoke the day before a conference on t...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4116</th>\n",
              "      <td>main-news</td>\n",
              "      <td>headlines</td>\n",
              "      <td>2013</td>\n",
              "      <td>453</td>\n",
              "      <td>2.4</td>\n",
              "      <td>Clashes as Greeks protest against more cutbacks</td>\n",
              "      <td>Clashes erupt as Greek vote nears</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "           genre  ...                                         sentence_2\n",
              "2340  main-forum  ...      buwhahahaha Breathe In, Breathe Out, Move On.\n",
              "4445   main-news  ...  Bangladesh building-collapse death toll climbs...\n",
              "4300   main-news  ...   Navy jet crashes into Virginia apartments,9 hurt\n",
              "3474   main-news  ...  antonov spoke the day before a conference on t...\n",
              "4116   main-news  ...                  Clashes erupt as Greek vote nears\n",
              "\n",
              "[5 rows x 7 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vIBUF4k4jPfR",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "n_wq1Y2fjhNX",
        "colab_type": "text"
      },
      "source": [
        "# **Text Preprocessing**\n",
        "Prepare text data according to Electra format as same as BERT. We have to tokenize the text sequences/sentences as per the Electra requirements before feeding them into Model. For this, we will use ElectraTokenizer from hugging face."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3l3vNjfmXtn6",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 615
        },
        "outputId": "6319cae6-d457-438c-e961-498819c52ea4"
      },
      "source": [
        "#tranformer library from Hugging face\n",
        "!pip install transformers"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting transformers\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/27/3c/91ed8f5c4e7ef3227b4119200fc0ed4b4fd965b1f0172021c25701087825/transformers-3.0.2-py3-none-any.whl (769kB)\n",
            "\r\u001b[K     |▍                               | 10kB 22.2MB/s eta 0:00:01\r\u001b[K     |▉                               | 20kB 6.3MB/s eta 0:00:01\r\u001b[K     |█▎                              | 30kB 6.3MB/s eta 0:00:01\r\u001b[K     |█▊                              | 40kB 7.0MB/s eta 0:00:01\r\u001b[K     |██▏                             | 51kB 7.4MB/s eta 0:00:01\r\u001b[K     |██▋                             | 61kB 8.4MB/s eta 0:00:01\r\u001b[K     |███                             | 71kB 8.4MB/s eta 0:00:01\r\u001b[K     |███▍                            | 81kB 8.5MB/s eta 0:00:01\r\u001b[K     |███▉                            | 92kB 8.9MB/s eta 0:00:01\r\u001b[K     |████▎                           | 102kB 9.2MB/s eta 0:00:01\r\u001b[K     |████▊                           | 112kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████▏                          | 122kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████▌                          | 133kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████                          | 143kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████▍                         | 153kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████▉                         | 163kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████▎                        | 174kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████▊                        | 184kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████                        | 194kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████▌                       | 204kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████                       | 215kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████▍                      | 225kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████▉                      | 235kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████▎                     | 245kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████▋                     | 256kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████                     | 266kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████▌                    | 276kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████                    | 286kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████▍                   | 296kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████▉                   | 307kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████▏                  | 317kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████▋                  | 327kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████                  | 337kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████▌                 | 348kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████                 | 358kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████▍                | 368kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████▊                | 378kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████▏               | 389kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████▋               | 399kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████████               | 409kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████████▌              | 419kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████              | 430kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████▎             | 440kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████▊             | 450kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████████▏            | 460kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████████▋            | 471kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████████            | 481kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████████▌           | 491kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████████▉           | 501kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████▎          | 512kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████▊          | 522kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████████▏         | 532kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████████▋         | 542kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████████████         | 552kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████████████▍        | 563kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████████████▉        | 573kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████████████▎       | 583kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████████████▊       | 593kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▏      | 604kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▋      | 614kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████████████      | 624kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████████████▍     | 634kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████████████▉     | 645kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▎    | 655kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▊    | 665kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████████████████▏   | 675kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████████████████▌   | 686kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████   | 696kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████▍  | 706kB 9.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████▉  | 716kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████▎ | 727kB 9.2MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████▊ | 737kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████ | 747kB 9.2MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████▌| 757kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 768kB 9.2MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 778kB 9.2MB/s \n",
            "\u001b[?25hRequirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from transformers) (1.18.5)\n",
            "Requirement already satisfied: tqdm>=4.27 in /usr/local/lib/python3.6/dist-packages (from transformers) (4.41.1)\n",
            "Collecting tokenizers==0.8.1.rc1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/40/d0/30d5f8d221a0ed981a186c8eb986ce1c94e3a6e87f994eae9f4aa5250217/tokenizers-0.8.1rc1-cp36-cp36m-manylinux1_x86_64.whl (3.0MB)\n",
            "\u001b[K     |████████████████████████████████| 3.0MB 30.5MB/s \n",
            "\u001b[?25hRequirement already satisfied: filelock in /usr/local/lib/python3.6/dist-packages (from transformers) (3.0.12)\n",
            "Requirement already satisfied: dataclasses; python_version < \"3.7\" in /usr/local/lib/python3.6/dist-packages (from transformers) (0.7)\n",
            "Collecting sentencepiece!=0.1.92\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/d4/a4/d0a884c4300004a78cca907a6ff9a5e9fe4f090f5d95ab341c53d28cbc58/sentencepiece-0.1.91-cp36-cp36m-manylinux1_x86_64.whl (1.1MB)\n",
            "\u001b[K     |████████████████████████████████| 1.1MB 57.7MB/s \n",
            "\u001b[?25hCollecting sacremoses\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/7d/34/09d19aff26edcc8eb2a01bed8e98f13a1537005d31e95233fd48216eed10/sacremoses-0.0.43.tar.gz (883kB)\n",
            "\u001b[K     |████████████████████████████████| 890kB 54.5MB/s \n",
            "\u001b[?25hRequirement already satisfied: requests in /usr/local/lib/python3.6/dist-packages (from transformers) (2.23.0)\n",
            "Requirement already satisfied: packaging in /usr/local/lib/python3.6/dist-packages (from transformers) (20.4)\n",
            "Requirement already satisfied: regex!=2019.12.17 in /usr/local/lib/python3.6/dist-packages (from transformers) (2019.12.20)\n",
            "Requirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from sacremoses->transformers) (1.15.0)\n",
            "Requirement already satisfied: click in /usr/local/lib/python3.6/dist-packages (from sacremoses->transformers) (7.1.2)\n",
            "Requirement already satisfied: joblib in /usr/local/lib/python3.6/dist-packages (from sacremoses->transformers) (0.16.0)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.6/dist-packages (from requests->transformers) (2020.6.20)\n",
            "Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.6/dist-packages (from requests->transformers) (2.10)\n",
            "Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.6/dist-packages (from requests->transformers) (3.0.4)\n",
            "Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.6/dist-packages (from requests->transformers) (1.24.3)\n",
            "Requirement already satisfied: pyparsing>=2.0.2 in /usr/local/lib/python3.6/dist-packages (from packaging->transformers) (2.4.7)\n",
            "Building wheels for collected packages: sacremoses\n",
            "  Building wheel for sacremoses (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for sacremoses: filename=sacremoses-0.0.43-cp36-none-any.whl size=893260 sha256=2bcfaf1d57a982e145ecf5bf44540584f49a5f4dfa5fab4b39338702c17544f1\n",
            "  Stored in directory: /root/.cache/pip/wheels/29/3c/fd/7ce5c3f0666dab31a50123635e6fb5e19ceb42ce38d4e58f45\n",
            "Successfully built sacremoses\n",
            "Installing collected packages: tokenizers, sentencepiece, sacremoses, transformers\n",
            "Successfully installed sacremoses-0.0.43 sentencepiece-0.1.91 tokenizers-0.8.1rc1 transformers-3.0.2\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tFQycZnUZPQI",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 66,
          "referenced_widgets": [
            "5b16ffff66c14c94972d877381c3616e",
            "66fef79997c24c70bbf172405b437210",
            "f4e474becb154483ab4b127ffa98af0b",
            "1c60f452c0e643d792b0b86e4dfe61e2",
            "0f197cf5a5694dd4858cf6925acba372",
            "42bba2c009e94aa2851c2f5b02467b0b",
            "5230bca55323461b8f7f6446c3b8dc17",
            "8745a7d59ac94df9bbdf297ab13733ec"
          ]
        },
        "outputId": "dc333b5d-7133-4946-895a-98d3516cc8e5"
      },
      "source": [
        "#import libraries from hugging face\n",
        "from transformers import ElectraModel, ElectraTokenizer, BertTokenizer, BertModel\n",
        "tokenizer= ElectraTokenizer.from_pretrained('google/electra-base-discriminator') #initiate tokenizer "
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "5b16ffff66c14c94972d877381c3616e",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=231508.0, style=ProgressStyle(descripti…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Wg8A1suSpg_3",
        "colab_type": "text"
      },
      "source": [
        "# Hugging Face Tokenizer\n",
        "Example of  tokenizer "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V1iP_RrDnTvn",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "2d6f6ad4-bb8d-4a8c-fdd7-7918dbf4e160"
      },
      "source": [
        "#text\n",
        "text = df_train.sentence_1[0]\n",
        "print('Raw text: {}'.format(text))\n",
        "\n",
        "#Tokenize the text\n",
        "tokenize = tokenizer.tokenize(text)\n",
        "print('Tokenized text: {}'.format(tokenize))\n",
        "\n",
        "#encode text\n",
        "input_ids =tokenizer.encode(text) # add '[CLS]' and '[SEP]' tokens\n",
        "print('Encoded text: {}'.format(input_ids))\n",
        "\n",
        "#convert input_ids back to tokens\n",
        "ids_to_tokens= tokenizer.convert_ids_to_tokens(input_ids)\n",
        "print('ids_to_tokens: {}'.format(ids_to_tokens))\n",
        "\n",
        "#convert tokens back to input_ids\n",
        "ids_to_tokens= tokenizer.convert_tokens_to_ids(ids_to_tokens)\n",
        "print('ids_to_token: {}'.format(ids_to_tokens))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Raw text: A plane is taking off.\n",
            "Tokenized text: ['a', 'plane', 'is', 'taking', 'off', '.']\n",
            "Encoded text: [101, 1037, 4946, 2003, 2635, 2125, 1012, 102]\n",
            "ids_to_tokens: ['[CLS]', 'a', 'plane', 'is', 'taking', 'off', '.', '[SEP]']\n",
            "ids_to_token: [101, 1037, 4946, 2003, 2635, 2125, 1012, 102]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ROj6x33Hnwcn",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "m23okuXjv0Vv",
        "colab_type": "text"
      },
      "source": [
        "#Count tokens in Sequences\n",
        "Since we have to have equal length of every sequence, check the maximun token length in every sentences in all datasets and  we will use this max length for all sequences as fix-length size while doing sequence padding.\n",
        "\n",
        "Or, we may use the other  best max-length. So that there is no risk of loss of much tokens while doing truncating and not require much of padding  in the sequences."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nom7ekL4nwZ8",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 187
        },
        "outputId": "b5554eb8-f67d-4537-cbe8-174f6aa7a382"
      },
      "source": [
        "#counts tokens in sequences\n",
        "dataset_all = [df_train.sentence_1, df_train.sentence_2, df_dev.sentence_1, df_dev.sentence_2,df_test.sentence_1,df_test.sentence_2] #list of all dataset\n",
        "dataset_type_name = ['df_train.sentence_1', 'df_train.sentence_2', 'df_dev.sentence_1', 'df_dev.sentence_2', 'df_test.sentence_1', 'df_test.sentence_2'] #name of dataset\n",
        "\n",
        "max_len = 0 # tokens max count in overall datasets\n",
        "token_len_list = [] #store token counts for each dataset\n",
        "higest_token_len = {} # store higest token counts in each dataset\n",
        "\n",
        "\n",
        "#zip and iterate over all datasets\n",
        "for nam_data, each_dataset in zip(dataset_type_name, dataset_all):\n",
        "  token_len_dataset = [] #store each sequence count of each dataset\n",
        "  max_len_seq= 0 # max count of  sequence in each dataset \n",
        "\n",
        "  #iterate over sequences in each dataset\n",
        "  for sent in each_dataset:\n",
        "    tokens = tokenizer.encode(sent) #tokenize\n",
        "    tokens_len = len(tokens) # count length\n",
        "    token_len_dataset.append(tokens_len) \n",
        "    \n",
        "    \n",
        "    # keep tracking the higest counts in each dataset\n",
        "    if tokens_len > max_len_seq:\n",
        "      max_len_seq = tokens_len\n",
        "    \n",
        "    #higest count overall dataset\n",
        "    if max_len_seq > max_len:\n",
        "      max_len = max_len_seq\n",
        "       \n",
        "  higest_token_len[nam_data] =  max_len_seq \n",
        "  token_len_list.append(token_len_dataset)\n",
        "\n",
        "print(\"Higest tokens number in 3 datasets: %s .\" % (max_len))\n",
        "print()\n",
        "print('Higest counts in each dataset:\\n')\n",
        "higest_token_len\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Higest tokens number in 3 datasets: 70 .\n",
            "\n",
            "Higest counts in each dataset:\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'df_dev.sentence_1': 45,\n",
              " 'df_dev.sentence_2': 53,\n",
              " 'df_test.sentence_1': 43,\n",
              " 'df_test.sentence_2': 46,\n",
              " 'df_train.sentence_1': 70,\n",
              " 'df_train.sentence_2': 63}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jG-zdvMRv_T-",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 788
        },
        "outputId": "d533353a-61d7-4083-bd7b-7f54daad210b"
      },
      "source": [
        "#Plot the diagram to show the tokens' length distribution in dataset\n",
        "import seaborn as sns\n",
        "from matplotlib import pyplot as plt\n",
        "\n",
        "# data to plot\n",
        "data = token_len_list \n",
        "\n",
        "#creates subplots  with nrows=3, ncols=2\n",
        "fig, axes = plt.subplots(nrows=3, ncols=2, figsize=(12,11))\n",
        "sns.set(style='darkgrid') #set plot style\n",
        "\n",
        "#change 'axes' to 1d, zip and emumerate\n",
        "for i, (name, axe)  in enumerate(zip(dataset_type_name, axes.flatten())):\n",
        "  sns.distplot(token_len_list[i], axlabel ='Tokens Count', ax=axe) #distplot type\n",
        "  name = name + ': (tokens_max_count= '+ str(higest_token_len[name]) + ')' #title (for example: df_train.sentence_1:(max_count=70))\n",
        "  axe.set_title(name) #title for each suplots\n",
        "fig.tight_layout()#fit nicely\n",
        "plt.show()\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x792 with 6 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Y1Kpor7awXk6",
        "colab_type": "text"
      },
      "source": [
        "**Observation**:  The max tokens count  in a sequence is 70 in all dataset. Among all dataset, the train dataset(both in sentence1 and sentence 2) have higer tokens count of 70 and 63 respectively. But, the majority of tokens length are about 40 counts. \n",
        "\n",
        "So, this gives us the good idea for selecting the max-length of tokens in a sequence.  We may put our max-length of 60, so that we will not truncate more tokens(loss information) and also don't have to do more padding.\n",
        "\n",
        "But, we may use 120 tokens length (double of a single sequence) for the model architecure where a pair of sentence are required to feed together."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "StH2fjqtwcD3",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bmAcWXS6_UP5",
        "colab_type": "text"
      },
      "source": [
        "# Part-1: Sentence Level Embedding Extraction\n",
        "For Sentence Embedding Extraction, last 4 hidden layers are concatenated and mean pooling of words in a sequence/sentence is done. So, the final dimension of a sentence vector representation is 768*4 =3072."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VbHtY2_Ta2aC",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#function to extract sentence level embedding from Bert Model\n",
        "def sent_embedding(sent, model):\n",
        "  \"\"\"\n",
        "   Concatenate from last 4 layers and Mean pooling of words in sentences\n",
        "\n",
        "  Args:\n",
        "     sent: list of sentences\n",
        "     model: BERT pre-trained model\n",
        "\n",
        "  Returns:\n",
        "     sent_embedding: 2d tensor \n",
        "\n",
        "  \"\"\"\n",
        "  # tokenize  each sequence into fix-length size of 70 tokens\n",
        "  tokens = tokenizer.batch_encode_plus( sent,\n",
        "                                       max_length= 70, # each sequence size\n",
        "                                       padding= 'max_length', # padding\n",
        "                                       add_special_tokens = True, # add '[CLS]'and '[SEP]'\n",
        "                                       return_attention_mask = True, # attention mask \n",
        "                                       return_tensors='pt',  # return PyTorch tensors\n",
        "                                      )\n",
        "   \n",
        "  \n",
        "\n",
        "  #feed input_ids to BERT model\n",
        "  with torch.no_grad():  # reduces memory consumption\n",
        "        outputs = model(tokens['input_ids'], tokens['attention_mask'] ) # feed input_ids, attention_mask\n",
        "        hidden_states = outputs[1] # all hidden layers\n",
        "\n",
        "  concat_embeding = torch.cat(tuple([hidden_states[i] for i in [-4,-3,-2,-1]]), dim = 2) #concatenate last 4 hidden layers\n",
        "  print('Dimension after Last_four_layers concatenation: ',concat_embeding.shape )\n",
        "  padded = tokens['attention_mask'].unsqueeze(2) # insert size one at 2 position\n",
        "  mul_out = torch.mul(concat_embeding, padded)# make zeros vector for paddding tokens\n",
        "  sent_embeding = mul_out.mean(dim=1) # average pooling  of tokens\n",
        "  print('Sentence embedding dimension: ',sent_embeding.shape )\n",
        "   \n",
        "  return sent_embeding\n",
        "\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WlPSjJWFahXI",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000,
          "referenced_widgets": [
            "ce3983cca06142d49cbbf101b28cf3d9",
            "678f946a99af41968a18db63a841e0e9",
            "6edf8b8caef149ab98271915a5126eac",
            "b75609f89e264b059eb0a5f9a46cc7f6",
            "fc4c682557204486b130ab67985ef32c",
            "b06674f688a4438e8f7d11dd137ffe94",
            "950bb8f2c0794397a5125b3819e4a43d",
            "c2228155806d4bf6b8f59d2e5f251e91",
            "3957196b6c8b4572935d00dd78b3b056",
            "8055d0ac75654a5ea8ebc1e712cf0ef1",
            "3036b3bd4a4d4d1891530f8df259317e",
            "863b14e4780c4452bd911dfcddfe4219",
            "3acd58850e954797b9f87811504096e5",
            "36591ad71cec45138c172aadfbd5d157",
            "2261e2cd10f74c75ae19890db5a9b1d2",
            "6669284c25a64c79abfc63fd0b7377f2"
          ]
        },
        "outputId": "8fb51b64-6b27-4de9-9426-87eeffa200be"
      },
      "source": [
        "# initiate the pre-trained model\n",
        "model = ElectraModel.from_pretrained( 'google/electra-base-discriminator',\n",
        "                                     output_hidden_states = True # return all hidden-states\n",
        "                                    )\n",
        "#put model in evaluation\n",
        "model.eval()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "ce3983cca06142d49cbbf101b28cf3d9",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=467.0, style=ProgressStyle(description_…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "3957196b6c8b4572935d00dd78b3b056",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=440343552.0, style=ProgressStyle(descri…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Some weights of the model checkpoint at google/electra-base-discriminator were not used when initializing ElectraModel: ['electra.embeddings_project.weight', 'electra.embeddings_project.bias']\n",
            "- This IS expected if you are initializing ElectraModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPretraining model).\n",
            "- This IS NOT expected if you are initializing ElectraModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "ElectraModel(\n",
              "  (embeddings): ElectraEmbeddings(\n",
              "    (word_embeddings): Embedding(30522, 768, padding_idx=0)\n",
              "    (position_embeddings): Embedding(512, 768)\n",
              "    (token_type_embeddings): Embedding(2, 768)\n",
              "    (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "    (dropout): Dropout(p=0.1, inplace=False)\n",
              "  )\n",
              "  (encoder): BertEncoder(\n",
              "    (layer): ModuleList(\n",
              "      (0): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (1): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (2): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (3): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (4): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (5): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (6): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (7): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (8): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (9): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (10): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "      (11): BertLayer(\n",
              "        (attention): BertAttention(\n",
              "          (self): BertSelfAttention(\n",
              "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "          (output): BertSelfOutput(\n",
              "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
              "            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "            (dropout): Dropout(p=0.1, inplace=False)\n",
              "          )\n",
              "        )\n",
              "        (intermediate): BertIntermediate(\n",
              "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
              "        )\n",
              "        (output): BertOutput(\n",
              "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
              "          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
              "          (dropout): Dropout(p=0.1, inplace=False)\n",
              "        )\n",
              "      )\n",
              "    )\n",
              "  )\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4eYGVHu1sOl-",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 442
        },
        "outputId": "76d962bb-fd38-4711-ef81-9edbc36aaacd"
      },
      "source": [
        "#check paramerts in model\n",
        "params = list(model.named_parameters())\n",
        "#len(model.state_dict())\n",
        "print('total different named paramerters: {}'.format(len(params)))\n",
        "print()\n",
        "for p in params[0:5]:\n",
        "    print(\"{:<55} {:>12}\".format(p[0], str(tuple(p[1].size()))))\n",
        "\n",
        "print()\n",
        "print()\n",
        "for p in params[-16:]:\n",
        "    print(\"{:<55} {:>12}\".format(p[0], str(tuple(p[1].size()))))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "total different named paramerters: 197\n",
            "\n",
            "embeddings.word_embeddings.weight                       (30522, 768)\n",
            "embeddings.position_embeddings.weight                     (512, 768)\n",
            "embeddings.token_type_embeddings.weight                     (2, 768)\n",
            "embeddings.LayerNorm.weight                                   (768,)\n",
            "embeddings.LayerNorm.bias                                     (768,)\n",
            "\n",
            "\n",
            "encoder.layer.11.attention.self.query.weight              (768, 768)\n",
            "encoder.layer.11.attention.self.query.bias                    (768,)\n",
            "encoder.layer.11.attention.self.key.weight                (768, 768)\n",
            "encoder.layer.11.attention.self.key.bias                      (768,)\n",
            "encoder.layer.11.attention.self.value.weight              (768, 768)\n",
            "encoder.layer.11.attention.self.value.bias                    (768,)\n",
            "encoder.layer.11.attention.output.dense.weight            (768, 768)\n",
            "encoder.layer.11.attention.output.dense.bias                  (768,)\n",
            "encoder.layer.11.attention.output.LayerNorm.weight            (768,)\n",
            "encoder.layer.11.attention.output.LayerNorm.bias              (768,)\n",
            "encoder.layer.11.intermediate.dense.weight               (3072, 768)\n",
            "encoder.layer.11.intermediate.dense.bias                     (3072,)\n",
            "encoder.layer.11.output.dense.weight                     (768, 3072)\n",
            "encoder.layer.11.output.dense.bias                            (768,)\n",
            "encoder.layer.11.output.LayerNorm.weight                      (768,)\n",
            "encoder.layer.11.output.LayerNorm.bias                        (768,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "88eYimtbxApr",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PADp4hO9_uhO",
        "colab_type": "text"
      },
      "source": [
        "# Extract Sentence embedding\n",
        "Now, lets extract sentence level vector representation for all datasets."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8aaghiCIsRYl",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 119
        },
        "outputId": "a19f5d3a-9dc3-4a9d-d038-887b03d5cf62"
      },
      "source": [
        "%%time\n",
        "#dev dataset\n",
        "df_dev_sentence_1 = sent_embedding(df_dev.sentence_1, model)\n",
        "df_dev_sentence_2 = sent_embedding(df_dev.sentence_2, model)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Dimension after Last_four_layers concatenation:  torch.Size([1500, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1500, 3072])\n",
            "Dimension after Last_four_layers concatenation:  torch.Size([1500, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1500, 3072])\n",
            "CPU times: user 10min 10s, sys: 8.42 s, total: 10min 18s\n",
            "Wall time: 10min 19s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "laHXhy-7sVt5",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 119
        },
        "outputId": "08052f41-fa7a-4d1a-cba2-d681f0750eb4"
      },
      "source": [
        "%%time\n",
        "#test dataset\n",
        "df_test_sentence_1 = sent_embedding(df_test.sentence_1, model)\n",
        "df_test_sentence_2 = sent_embedding(df_test.sentence_2, model)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Dimension after Last_four_layers concatenation:  torch.Size([1379, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1379, 3072])\n",
            "Dimension after Last_four_layers concatenation:  torch.Size([1379, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1379, 3072])\n",
            "CPU times: user 9min 35s, sys: 2.29 s, total: 9min 37s\n",
            "Wall time: 9min 38s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "S7Z5qnojyu9Q",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 323
        },
        "outputId": "c025aab7-4cd2-4001-afc2-1418ecd1b288"
      },
      "source": [
        "%%time\n",
        "#train dataset\n",
        "train_embed_sentence_1 =  []\n",
        "train_embed_sentence_2 =  []\n",
        "\n",
        "#make batches\n",
        "for x in range(0, df_train.shape[0], 1500):\n",
        "  embed_1= sent_embedding(df_train.sentence_1[x: x+1500], model)\n",
        "  embed_2= sent_embedding(df_train.sentence_2[x: x+1500], model)\n",
        "  train_embed_sentence_1.extend(embed_1)\n",
        "  train_embed_sentence_2.extend(embed_2)\n",
        "\n",
        "#sentence level vector represetnation in train dataset\n",
        "df_train_sentence_1 = torch.stack(train_embed_sentence_1) # change list of tensors >>> 2d tensor\n",
        "df_train_sentence_2 = torch.stack(train_embed_sentence_2) # change list of tensors >>> 2d tensor\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Dimension after Last_four_layers concatenation:  torch.Size([1500, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1500, 3072])\n",
            "Dimension after Last_four_layers concatenation:  torch.Size([1500, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1500, 3072])\n",
            "Dimension after Last_four_layers concatenation:  torch.Size([1500, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1500, 3072])\n",
            "Dimension after Last_four_layers concatenation:  torch.Size([1500, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1500, 3072])\n",
            "Dimension after Last_four_layers concatenation:  torch.Size([1500, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1500, 3072])\n",
            "Dimension after Last_four_layers concatenation:  torch.Size([1500, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1500, 3072])\n",
            "Dimension after Last_four_layers concatenation:  torch.Size([1249, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1249, 3072])\n",
            "Dimension after Last_four_layers concatenation:  torch.Size([1249, 70, 3072])\n",
            "Sentence embedding dimension:  torch.Size([1249, 3072])\n",
            "CPU times: user 39min 22s, sys: 9.74 s, total: 39min 31s\n",
            "Wall time: 39min 33s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fasnOCPPsddz",
        "colab_type": "text"
      },
      "source": [
        "# Calculate Cosine Similarity\n",
        "Now, sentence level vector representation of each pair sentence is used to find the cosine similarity. Cosine similarity is calculated on all 'dev' and 'test' datasets."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EBP8ADEusgJw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from scipy.spatial.distance import cosine #cosine distance\n",
        "from sklearn.preprocessing import MinMaxScaler # minmax scaling\n",
        "from scipy.stats import pearsonr # pearson correlation\n",
        "from scipy.stats import spearmanr # spearman correlation\n",
        "from collections import defaultdict #store scores"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BcvdIlgGsniC",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#function to calculate similarity score \n",
        "def calc_similarity_score(embed_1, embed_2):\n",
        "  \"\"\"\n",
        "  Args:\n",
        "    embed_1(numpy array): first sentence embedding\n",
        "    embed_2(numpy array): second sentence embedding\n",
        "\n",
        "  Return:\n",
        "    scaled_cosine_score(list) : cosine score between embed_1 and embed_2 \n",
        "    with range (0,5)\n",
        "\n",
        "\n",
        "  \"\"\"\n",
        "  \n",
        "  #collect similarity score\n",
        "  cosine_sim_score = []\n",
        "  \n",
        "  for x,y in zip(embed_1, embed_2):\n",
        "\n",
        "    distance = cosine(x, y) # cosine distance\n",
        "    cosine_sim = 1 - distance # cosine similarity\n",
        "    cosine_sim_score.append(cosine_sim) \n",
        "\n",
        "  \n",
        "\n",
        "\n",
        "  #scaled the similarity score to (0,5)\n",
        "  scaler = MinMaxScaler(feature_range=(0,5)) # initiate minmaxscaler\n",
        "  scaled_cosine_score = scaler.fit_transform(np.transpose([cosine_sim_score])) #scaled to (0,5)\n",
        "  \n",
        "  return scaled_cosine_score.round(3)\n",
        "  "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pTPZtZu_xdrO",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 272
        },
        "outputId": "385c6e3e-1c50-48cc-c446-5e23f8a03e65"
      },
      "source": [
        "#Pearson and Spearman score\n",
        "first_sent_embd = [df_train_sentence_1, df_dev_sentence_1, df_test_sentence_1] #first sentence embedding\n",
        "second_sent_embd = [df_train_sentence_2, df_dev_sentence_2, df_test_sentence_2] #second sentence embedding\n",
        "gold_labels = [df_train.loc[:,'score(0-5)'].values, df_dev.loc[:,'score(0-5)'].values, df_test.loc[:,'score(0-5)'].values ] #gold scores\n",
        "data_type = ['Train', 'Validation', 'Test'] #datasets names\n",
        "\n",
        "pear_spear_score = defaultdict(list) #store scores\n",
        "\n",
        "#zip and enumerate 'first_sent_embd' and 'second_sent_embd'\n",
        "for i, (first_sent, second_sent) in enumerate(zip(first_sent_embd,second_sent_embd)):\n",
        "  cosine_sim_score = calc_similarity_score(first_sent, second_sent) #cosine similarity score\n",
        "  cosine_sim_score = cosine_sim_score.flatten()# change to 1d\n",
        "\n",
        "  person_score, _ = pearsonr(gold_labels[i], cosine_sim_score) #pearson score\n",
        "  pear_spear_score['pearson'].append(person_score)\n",
        "  \n",
        "\n",
        "  sperman_score, _ = spearmanr(gold_labels[i], cosine_sim_score)#spearman score\n",
        "  pear_spear_score['spearman'].append(sperman_score)\n",
        "\n",
        "  print(data_type[i], 'data: ')\n",
        "  print('=='*6)\n",
        "  print(\"Pearson score: {} \\nSpearman score: {}\".format(person_score, sperman_score))\n",
        "  print()\n",
        "  \n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train data: \n",
            "============\n",
            "Pearson score: 0.3521128219242041 \n",
            "Spearman score: 0.354980844369472\n",
            "\n",
            "Validation data: \n",
            "============\n",
            "Pearson score: 0.4676935776575393 \n",
            "Spearman score: 0.5036069569686787\n",
            "\n",
            "Test data: \n",
            "============\n",
            "Pearson score: 0.31468780026124793 \n",
            "Spearman score: 0.3382698422445761\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "y1lIpMQzCpem",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "O1GHAFYbxeGi",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 111
        },
        "outputId": "3552564e-343a-4321-9e18-387c749b6132"
      },
      "source": [
        "#show in panda dataframe\n",
        "df_score = pd.DataFrame.from_dict(pear_spear_score,\n",
        "                          orient='index',\n",
        "                          columns=['Train', 'Dev', 'Test']\n",
        "                          )\n",
        "df_score.drop('Train', 1) #only train and test \n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Dev</th>\n",
              "      <th>Test</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>pearson</th>\n",
              "      <td>0.467694</td>\n",
              "      <td>0.314688</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>spearman</th>\n",
              "      <td>0.503607</td>\n",
              "      <td>0.338270</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "               Dev      Test\n",
              "pearson   0.467694  0.314688\n",
              "spearman  0.503607  0.338270"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 49
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NV_sssxGGvnj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c90Bk95_xiA2",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ikNSWXvG6Q0S",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "upIBtMuw6VHL",
        "colab_type": "text"
      },
      "source": [
        "# Part-2: Fine-Tune ELECTRA\n",
        "The custom model is defined using pytorch nn.module. And, The pre-trained ELECTRA(electra-base-discriminator) model is fine-tunned with STS-b 'train' dataset and evulate on 'dev' test"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HIlP2NMr6Q7Z",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#import necessary libraries\n",
        "import torch.nn as nn\n",
        "from torch.utils.data import Dataset, DataLoader #pytorch dataset and dataloader\n",
        "from transformers import  AdamW, get_linear_schedule_with_warmup #optimizer and learning rate scheduler\n",
        "from collections import defaultdict  #store scores"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZLKG-hSh6mDp",
        "colab_type": "text"
      },
      "source": [
        "# Custom Model\n",
        "The two different network stuctures are tried  for fine-tuning the pre-trained model. They are as follows:\n",
        "\n",
        "1. Siamese Network: In this Siamese Network structure, two different sentences(a pair of sentences) are encoded separetely meaning they are encoded one by one, and their final vecotors are used to calculate the cosine similarity.  The output layer is just the cosine similairty. No additonal layers.  The vector representation of a sentence is the concatenation of the last 4 hidden layers, and mean pooling of the tokens in a inputed sequence. \n",
        "\n",
        "2. Regression Model: In this fine-tuning, the Linear Regression output layer is added on the top of Bert. The two sentences(a pair of sentences) are feeded and encoded together. And, the final output is the regression ouput.\n",
        "\n",
        "The Mean Square Error(MSE) Loss is used as loss function  for both the Netowrk structures."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "X6t4-irI6nh8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#define custom model with cosine-similarity output\n",
        "class ElectraSentenceSimilarityModelCos(nn.Module):\n",
        "  \"\"\"\n",
        "  Model extracts sentence embedding of a pair of sentences one by one and\n",
        "  return  cosine similarity between them.\n",
        "  \"\"\"\n",
        "  \n",
        "  def __init__(self):\n",
        "    super().__init__()\n",
        "    self.electra = ElectraModel.from_pretrained( 'google/electra-base-discriminator',\n",
        "                                     output_hidden_states = True) # all hidden-states\n",
        "    self.cosine_sim = nn.CosineSimilarity(dim=1, eps=1e-6)#cosine similarity\n",
        "   \n",
        "  #feed a pair of sentences' input_ids and attention_ids \n",
        "  def forward(self, input_ids_1, atten_ids_1, input_ids_2, atten_ids_2):\n",
        "    \"\"\"\n",
        "        Args:\n",
        "          input_ids_1(tensor) : input_ids of first sentence\n",
        "          input_ids_2(tensor) : input_ids of second sentence\n",
        "          atten_ids_1(tensor): attention_mask of first sentence\n",
        "          atten_ids_2(tensor): attention_mask of second sentence\n",
        "       Returns:\n",
        "          cosine_score(2d tensor): Cosine similarity between a pair of sentences\n",
        "    \"\"\"\n",
        "    \n",
        "    #first sentence embedding\n",
        "    output_1 = self.electra(input_ids = input_ids_1, attention_mask = atten_ids_1)\n",
        "    hidden_states_1 = output_1[1] # all layers hidden-states\n",
        "    concat_embeding_1 = torch.cat(tuple(hidden_states_1[i] for i in [-4,-3,-2,-1]), dim = 2) # concatenate last four hidden layers\n",
        "    padded_1 = atten_ids_1.unsqueeze(2) #insert size one at position 2\n",
        "    mul_out_1 = torch.mul(concat_embeding_1, padded_1)#make zeros vector for paddding tokens\n",
        "    sent_embeding_1 = mul_out_1.mean(dim=1) # pooling average of tokens\n",
        "  \n",
        "    #second sentence embedding\n",
        "    output_2 = self.electra(input_ids = input_ids_2,attention_mask = atten_ids_2)\n",
        "    hidden_states_2 = output_2[1]\n",
        "    concat_embeding_2 = torch.cat(tuple(hidden_states_2[i] for i in [-4,-3,-2,-1]), dim = 2) # last four layers\n",
        "    padded_2 = atten_ids_2.unsqueeze(2) \n",
        "    mul_out_2 = torch.mul(concat_embeding_2, padded_2)\n",
        "    sent_embeding_2 = mul_out_2.mean(dim=1)\n",
        "\n",
        "    #calculate cosine similarity\n",
        "    cosine_score = self.cosine_sim(sent_embeding_1, sent_embeding_2)\n",
        "    return cosine_score\n",
        "    \n",
        "  "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1gRZvZhBymEh",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#define custom model with Regression output\n",
        "class ElectraSentenceSimilarityModelReg(nn.Module):\n",
        "  \"\"\"This model takes a pair of sentences at a time and returns regression output \"\"\"\n",
        "  \n",
        "  def __init__(self):\n",
        "    super().__init__()\n",
        "    self.electra = ElectraModel.from_pretrained( 'google/electra-base-discriminator',\n",
        "                                     output_hidden_states = True) # all hidden-states\n",
        "    self.drop = nn.Dropout(p=0.1) #droupout\n",
        "    self.regression = nn.Linear(self.electra.config.hidden_size,1) #output regresion layer\n",
        "      \n",
        "  #forward pass\n",
        "  def forward(self, input_ids, attention_masks, token_type_ids):\n",
        "    \"\"\"\n",
        "      Args:\n",
        "          input_ids_1(tensor) : input_ids of a pair of sentence\n",
        "          attention_mask_1(tensor): attention_mask of a pair of sentence\n",
        "\n",
        "       Returns:\n",
        "          regre_output(1d tensor) : regression output\n",
        "    \"\"\"\n",
        "  \n",
        "    #initiate pre-trained model\n",
        "    electra_output = self.electra( input_ids = input_ids, attention_mask = attention_masks,token_type_ids = token_type_ids)\n",
        "    #hidden_state = electra_output[0]# last_hidden_state\n",
        "    hidden_state = electra_output[0][:,0,:] #['CLS'] token of last hidden_state\n",
        "    drop_out =  self.drop(hidden_state) \n",
        "    regre_output =  self.regression(drop_out)# apply linnear tranformation\n",
        "    return regre_output\n",
        "\n",
        "  \n",
        "    "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "E0eeTgX0wxnW",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#define custom model with Regression output\n",
        "class ElectraBertSentenceSimilarityModelReg(nn.Module):\n",
        "  \"\"\"\n",
        "  This model  is the combination of Bert and Electra models. It takes a pair of \n",
        "  sentences at a time and returns regression output \n",
        "  \"\"\"\n",
        "  \n",
        "  def __init__(self):\n",
        "    super().__init__()\n",
        "    self.electra = ElectraModel.from_pretrained( 'google/electra-base-discriminator',\n",
        "                                     output_hidden_states = True) # pretrained Electra\n",
        "    self.bert = BertModel.from_pretrained('bert-base-uncased', output_hidden_states = True )#pretrained bert \n",
        "    self.drop = nn.Dropout(p=0.1) #droupout\n",
        "    self.regression = nn.Linear( self.electra.config.hidden_size*2,1) #output regresion layer\n",
        "      \n",
        "  #forward pass\n",
        "  def forward(self, input_ids, attention_masks, token_type_ids):\n",
        "    \"\"\"\n",
        "      Args:\n",
        "          input_ids_1(tensor) : input_ids of a pair of sentence\n",
        "          attention_mask_1(tensor): attention_mask of a pair of sentence\n",
        "\n",
        "       Returns:\n",
        "          regre_output(1d tensor) : regression output\n",
        "    \"\"\"\n",
        "\n",
        "    #initiate pre-trained bert model\n",
        "    bert_output = self.bert( input_ids = input_ids, attention_mask = attention_masks,token_type_ids = token_type_ids)\n",
        "    hidden_state_bert = bert_output[0][:,0,:]# ['CLS']  last_hidden_state\n",
        "    #drop_out_br =  self.drop(hidden_state_bert) \n",
        "   \n",
        "    \n",
        "    #initiate pre-trained electra model\n",
        "    electra_output = self.electra( input_ids = input_ids, attention_mask = attention_masks,token_type_ids = token_type_ids)\n",
        "    #hidden_state = electra_output[0]# last_hidden_state\n",
        "    hidden_state_electra = electra_output[0][:,0,:] #['CLS'] token of last hidden_state\n",
        "    #drop_out_el =  self.drop(hidden_state_electra) \n",
        "\n",
        "    #concatenation  bert and electra embedding\n",
        "    #bert_electra = torch.cat((drop_out_br, drop_out_el), dim=1) \n",
        "    bert_electra = torch.cat((hidden_state_bert, bert_electra), dim=1) \n",
        "    bert_electra = self.drop(bert_electra)\n",
        "    regre_output =  self.regression(bert_electra)# apply linnear tranformation'''\n",
        "    return regre_output\n",
        "  \n",
        "    "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9wz-ZO4E3Y1N",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JQjZvLq-6z-v",
        "colab_type": "text"
      },
      "source": [
        "# Create Pytorch Dataset\n",
        "Pytorch dataset is created that returns input_ids, attention_mask, token_type_ids, and gold_score from each input sequence.\n",
        "\n",
        "Since  we are trying two different model network architectures for fine-tuning, the way of feeding the pair of sentences are different from one architecture to another. For example, in a regression model, we feed a pair of sentences together while in Siamese network model, we feed one sentence and take sentence level embedding and same goes to its pair sentence. \n",
        "\n",
        "Thus, a function  is written that returns the pytorch dataset which fit for both the model architectures."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7mM94Pmg610O",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#define function to pytorch Dataset\n",
        "class SimilarityDataset(Dataset):\n",
        "  \"\"\" \n",
        "  Pytorch dataset that returns input_ids, attension_mask a pair of sentences, and \n",
        "  gold_score  of STS-B dataset. Dataset is return based on the model type \n",
        "  i.e. either ('regresion' or 'cosine'). Default is Siamese network structure.\n",
        "  \"\"\"\n",
        "\n",
        "  def __init__(self, sent_1s, sent_2s, scores, tokenizer, max_len, model_type = 'cosine'):\n",
        "    \"\"\"\n",
        "     Args:\n",
        "       sent_1s(numpy array): Array of first sentences\n",
        "       sent_2s(numpy array): Array of second sentences\n",
        "       scores(float): Gold score (0-5)\n",
        "       tokenizer: Higging Face Bert Tokenizer\n",
        "       max_len(int): Maximun token length\n",
        "       model_type(str): type of model architecture to calculate \n",
        "       ouput(for example, 'regression' or 'cosine')\n",
        "      \n",
        "    \"\"\"\n",
        "    self.sent_1s = sent_1s\n",
        "    self.sent_2s = sent_2s\n",
        "    self.scores = scores\n",
        "    self.tokenizer = tokenizer\n",
        "    self.max_len = max_len\n",
        "    self.model_type = model_type\n",
        "\n",
        "\n",
        "  def __len__(self):\n",
        "    return len(self.sent_1s) #total no. of samples\n",
        "\n",
        "\n",
        "  def __getitem__(self, item):\n",
        "    first_sent = self.sent_1s[item]  #first sentencs\n",
        "    second_sent = self.sent_2s[item] #second sentence\n",
        "    score = self.scores[item]\n",
        "\n",
        "    #print('model_type is: {}'.format(self.model_type))\n",
        "    # model type is 'Siamese Network'\n",
        "    if self.model_type == 'cosine':\n",
        "\n",
        "      #encode first sentence\n",
        "      first_sent_encoding = self.tokenizer( first_sent,\n",
        "                                            add_special_tokens=True, #add ['CLS'] and ['SEP']\n",
        "                                            max_length=self.max_len, # set max len for padding\n",
        "                                            return_token_type_ids=False, #set false to segment_ids\n",
        "                                            padding='max_length', # padding\n",
        "                                            return_attention_mask=True,  # return attention mask\n",
        "                                            return_tensors='pt', # pytorch tensor\n",
        "                                            )\n",
        "      \n",
        "\n",
        "\n",
        "      #encode second sentence\n",
        "      second_sent_encoding = self.tokenizer( second_sent,\n",
        "                                            add_special_tokens=True,\n",
        "                                            max_length=self.max_len,\n",
        "                                            return_token_type_ids=False,\n",
        "                                            padding= 'max_length',\n",
        "                                            return_attention_mask=True,\n",
        "                                            return_tensors='pt',\n",
        "                                            )\n",
        "      \n",
        "      return {\n",
        "          'input_ids' :{ \n",
        "              'first_sent' : first_sent_encoding['input_ids'].flatten(),\n",
        "              'second_sent' : second_sent_encoding['input_ids'].flatten()},\n",
        "\n",
        "          'attention_ids' :{\n",
        "              'first_sent' :first_sent_encoding['attention_mask'].flatten(),\n",
        "              'second_sent' :second_sent_encoding['attention_mask'].flatten()},\n",
        "\n",
        "          'score' : torch.tensor(score)}\n",
        "\n",
        "\n",
        "\n",
        "    #if model's out is regression type\n",
        "    else:\n",
        "\n",
        "      #encode input sequences\n",
        "      sent_encoding = self.tokenizer(   first_sent, second_sent,\n",
        "                                        add_special_tokens=True,\n",
        "                                        max_length=self.max_len,\n",
        "                                        return_token_type_ids=True,\n",
        "                                        padding= 'max_length',\n",
        "                                        truncation = True,\n",
        "                                        return_attention_mask=True,\n",
        "                                        return_tensors='pt',\n",
        "                                      )\n",
        "      return {\n",
        "            'input_ids' : sent_encoding['input_ids'].flatten(),\n",
        "            'attention_ids' : sent_encoding['attention_mask'].flatten(),\n",
        "            'token_type_ids' : sent_encoding['token_type_ids'].flatten(),\n",
        "            'score' : torch.tensor(score),}\n",
        "\n",
        "    \n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "i5RFb-3T3UHe",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RPK2Mnn67TPP",
        "colab_type": "text"
      },
      "source": [
        "# Pytorch Data Loader\n",
        "It provides an iterable over the given dataset."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wKt6oJZX7X2L",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#define pytorch data loader\n",
        "def pytorch_data_loader(df, tokenizer, max_len, batch_size, model_type='cosine'):\n",
        "  \"\"\"\n",
        "    Agrs:\n",
        "       df(panda dataframe): dataset\n",
        "       tokenizer(Hugging face tokeniser): Bert Tokonizer\n",
        "       max_len(int): Maximun tokens length in a sequence/sentence\n",
        "       batch_size(int): Number of samples in a batch\n",
        "       model_type(str): type of model architecture to calculate \n",
        "       ouputs(for example, 'regression' or 'cosine')\n",
        "\n",
        "    Return:\n",
        "      Dataloader(pytorch dataloader): Dataset with given batch size\n",
        "  \"\"\"\n",
        "  dataset = SimilarityDataset(\n",
        "      sent_1s =df.sentence_1.to_numpy(), #first sequence\n",
        "      sent_2s =df.sentence_2.to_numpy(), #second sequence\n",
        "      scores =df.loc[:,'scaled_score(0-1)'].to_numpy(), #gold score\n",
        "      tokenizer=tokenizer, #hugging face tokenizer\n",
        "      max_len=max_len, #max length for padding\n",
        "      model_type = model_type # return dataset as per model type\n",
        "  )\n",
        "\n",
        "  return DataLoader( \n",
        "      dataset,\n",
        "      batch_size=batch_size,\n",
        "      shuffle= True\n",
        "  )"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OYdy9FCmEe7S",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1mpWgfw_EfuQ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#define model training function\n",
        "def model_train(model, dataloader, device, loss_fun, optimizer, scheduler, model_type):\n",
        "  \"\"\"\n",
        "     Args:\n",
        "       model: Custom model\n",
        "       dataloader: Pytorh data loader\n",
        "       device: device to compute\n",
        "       loss_fun:  loss function( MES loss)\n",
        "       optimizer: optimizer (Adamw)\n",
        "       scheduler: learning rate schedule\n",
        "       model_type(str): type of model architecture('regression' or 'cosine')\n",
        "\n",
        "    Returns:\n",
        "       training loss, and Pearson correlation between gold_score and predicted score\n",
        "       \n",
        "  \"\"\"\n",
        "  print('Model is training based on {} ouput!!!'.format(model_type))\n",
        "\n",
        "  #put model in training mode\n",
        "  model.train()\n",
        "\n",
        "  losses = [] # store loss from each batch\n",
        "  pearson_score = [] #pearson score from each batch\n",
        "\n",
        "  #Siamese Model type(Cosine output)\n",
        "  if model_type == 'cosine':\n",
        "  \n",
        "    #iterate over each batch of training data\n",
        "    for batch in dataloader:\n",
        "\n",
        "      #first sent\n",
        "      input_ids_1 = batch['input_ids']['first_sent'].to(device)\n",
        "      atten_ids_1= batch['attention_ids']['first_sent'].to(device)\n",
        "      #second sent\n",
        "      input_ids_2 = batch['input_ids']['second_sent'].to(device)\n",
        "      atten_ids_2 = batch['attention_ids']['second_sent'].to(device)\n",
        "\n",
        "      #gold score\n",
        "      gold_score = batch['score'].to(device).float() # change to float dtype as same as model output\n",
        "      \n",
        "      #clear previously calculated gradients\n",
        "      model.zero_grad()\n",
        "\n",
        "      #model forward pass\n",
        "      output = model(input_ids_1, atten_ids_1, input_ids_2, atten_ids_2) # output dtype --> float tenosr \n",
        "      \n",
        "      #Calculate Pearson Correlation score\n",
        "      p_score = pearsonr(output.detach().cpu(), gold_score.detach().cpu())\n",
        "      pearson_score.append(p_score[0])\n",
        "      \n",
        "      #perform loss\n",
        "      loss = loss_fun(output, gold_score)\n",
        "      losses.append(loss.item()) \n",
        "\n",
        "      loss.backward() # perform backward pass to calculate gradients\n",
        "      nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)# clip the norm of the gradients to 1.0.\n",
        "      optimizer.step() #update parameters\n",
        "      scheduler.step() #update learning rate\n",
        "      \n",
        "    return np.mean(losses), np.mean(pearson_score)\n",
        "\n",
        "  #Regression model type\n",
        "  elif model_type == 'regression':\n",
        "\n",
        "    #iterate over each batch of training data\n",
        "    for batch in dataloader:\n",
        "      \n",
        "      #Input sequence \n",
        "      input_ids = batch['input_ids'].to(device)\n",
        "      atten_ids = batch['attention_ids'].to(device)\n",
        "      tok_type_ids = batch['token_type_ids'].to(device)\n",
        "\n",
        "      #gold score\n",
        "      gold_score = batch['score'].to(device).float() # change to float dtype as same as model output\n",
        "      \n",
        "      #clear previously calculated gradients\n",
        "      model.zero_grad()\n",
        "\n",
        "      #model forward pass\n",
        "      output = model(input_ids, atten_ids, tok_type_ids) # output dtype --> float tenosr \n",
        "      output = output.flatten()\n",
        "      \n",
        "      #Calculate Pearson Correlation score\n",
        "      p_score = pearsonr(output.detach().cpu(), gold_score.detach().cpu())\n",
        "      pearson_score.append(p_score[0])\n",
        "      \n",
        "      #perform loss\n",
        "      loss = loss_fun(output, gold_score)\n",
        "      losses.append(loss.item()) \n",
        "\n",
        "      loss.backward() # perform backward pass to calculate gradients\n",
        "      nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)# clip the norm of the gradients to 1.0.\n",
        "      optimizer.step() #update parameters\n",
        "      scheduler.step() #update learning rate\n",
        "    return np.mean(losses), np.mean(pearson_score)\n",
        "\n",
        "  else:\n",
        "    print('Pleae specify the model type for training !!!')\n",
        "\n",
        "  \n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K7TwtESiEkAi",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# define evaluation function\n",
        "def model_eval(model, dataloader, device, loss_fun, model_type):\n",
        "  \"\"\"\n",
        "     Args:\n",
        "       model: Custom model\n",
        "       dataloader: pytorch data loader\n",
        "       device: device to compute\n",
        "       loss_fun:  loss function( MES loss)\n",
        "       \n",
        "    Returns:\n",
        "       Evaluation loss, and Pearson & Spearman correlation between gold_score and predicted score\n",
        "       \n",
        "  \"\"\"\n",
        "  #print('Model is validation based on {} ouput!!!'.format(model_type))\n",
        "\n",
        "  #put model in evaluation mode\n",
        "  model.eval()\n",
        "  losses = []\n",
        "  pearson_score = []\n",
        "  sparman_score = []\n",
        "  \n",
        "  #tell model not to compute or store gradients\n",
        "  with torch.no_grad():\n",
        "    \n",
        "    #Siamese model type(cosine)\n",
        "    if model_type == 'cosine':\n",
        "      \n",
        "      #each batch of evaluation data\n",
        "      for batch in dataloader:\n",
        "        #first sentence\n",
        "        input_ids_1 = batch['input_ids']['first_sent'].to(device)\n",
        "        atten_ids_1 = batch['attention_ids']['first_sent'].to(device)\n",
        "\n",
        "        #second sentence\n",
        "        input_ids_2 = batch['input_ids']['second_sent'].to(device)\n",
        "        atten_ids_2 = batch['attention_ids']['second_sent'].to(device)\n",
        "\n",
        "        #gold score\n",
        "        gold_score = batch['score'].to(device).float() # change to float dtype as same as model output\n",
        "\n",
        "        #model forward pass\n",
        "        model = model\n",
        "        output = model(input_ids_1, atten_ids_1, input_ids_2, atten_ids_2) # output type --> float tensor \n",
        "        output = output.flatten()\n",
        "\n",
        "        #perform loss\n",
        "        loss = loss_fun(output, gold_score)\n",
        "        losses.append(loss.item()) \n",
        "        \n",
        "        #Calculate Pearson Correlation score\n",
        "        p_score = pearsonr(output.detach().cpu(), gold_score.detach().cpu())\n",
        "        pearson_score.append(p_score[0])\n",
        "\n",
        "        #Calculate Spearman Correlation score\n",
        "        sp_score = spearmanr(output.detach().cpu(), gold_score.detach().cpu())\n",
        "        sparman_score.append(sp_score[0])\n",
        "      return np.mean(losses), np.mean(pearson_score), np.mean(sparman_score)\n",
        "    \n",
        "\n",
        "    #regression model type\n",
        "    elif model_type == 'regression':\n",
        "\n",
        "      #each batch of evaluation data\n",
        "      for batch in dataloader:\n",
        "        #first sentence\n",
        "        input_ids = batch['input_ids'].to(device)\n",
        "        atten_ids = batch['attention_ids'].to(device)\n",
        "        tok_type_ids = batch['token_type_ids'].to(device)\n",
        "\n",
        "        #gold score\n",
        "        gold_score = batch['score'].to(device).float() # change to float dtype as same as model output\n",
        "\n",
        "        #model forward pass\n",
        "        model = model\n",
        "        output = model(input_ids, atten_ids, tok_type_ids) # output type --> float tensor \n",
        "        output= output.flatten()\n",
        "        \n",
        "        #perform loss\n",
        "        loss = loss_fun(output, gold_score)\n",
        "        losses.append(loss.item()) \n",
        "        \n",
        "        #Calculate Pearson Correlation score\n",
        "        p_score = pearsonr(output.detach().cpu(), gold_score.detach().cpu())\n",
        "        pearson_score.append(p_score[0])\n",
        "\n",
        "        #Calculate Spearman Correlation score\n",
        "        sp_score = spearmanr(output.detach().cpu(), gold_score.detach().cpu())\n",
        "        sparman_score.append(sp_score[0])\n",
        "\n",
        "      return np.mean(losses), np.mean(pearson_score), np.mean(sparman_score)\n",
        "\n",
        "    else:\n",
        "      print('Please specify the model type for evaluation !!!')\n",
        "\n",
        "  \n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xlzBysrgEguH",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lzi2R2EX7chd",
        "colab_type": "text"
      },
      "source": [
        "# Re-scaled gold score from (0-5) to (0,1)\n",
        "Before making the Data to Pytorch Dataloader, lets re-scaled the 'gold score' to (0,1). This is done because to make the same range as the  cosine similarity score i.e.(0,1). So, it will be  easy to calculate the loss during the model training."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k-vLmDhkCdkU",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#re-scale to all datasets\n",
        "df_train['scaled_score(0-1)'] = MinMaxScaler(feature_range=(0,1)).fit_transform(df_train[['score(0-5)']])\n",
        "df_dev['scaled_score(0-1)'] = MinMaxScaler(feature_range=(0,1)).fit_transform(df_dev[['score(0-5)']])\n",
        "df_test['scaled_score(0-1)'] = MinMaxScaler(feature_range=(0,1)).fit_transform(df_test[['score(0-5)']])\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SDsoWPjW7Y-D",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Create data loader for Regression model\n",
        "BATCH_SIZE = 16\n",
        "train_data_loader_reg = pytorch_data_loader( df_train.sample(frac=1.0), tokenizer, 120, BATCH_SIZE, 'regression') # train data loader\n",
        "dev_data_loader_reg = pytorch_data_loader(df_dev, tokenizer, 120, BATCH_SIZE , 'regression', )# dev data loader\n",
        "test_data_loader_reg = pytorch_data_loader(df_test, tokenizer, 120, BATCH_SIZE, 'regression' )# test data loader"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kHCzCPv64FNS",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Create data loader for  cosine similarity output\n",
        "train_data_loader_cos = pytorch_data_loader( df_train.sample(frac=1.0), tokenizer, 70, BATCH_SIZE) # train data loader\n",
        "dev_data_loader_cos= pytorch_data_loader(df_dev, tokenizer, 70, BATCH_SIZE )# dev data loader\n",
        "test_data_loader_cos = pytorch_data_loader(df_test, tokenizer, 70, BATCH_SIZE)# test data loader\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XhJ2sNk07j_A",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 425
        },
        "outputId": "d6a2e6e5-9866-455e-e27c-890b8d2e10bb"
      },
      "source": [
        "#chaek batch of dataloader\n",
        "data = next(iter(train_data_loader_reg))\n",
        "data\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'attention_ids': tensor([[1, 1, 1,  ..., 0, 0, 0],\n",
              "         [1, 1, 1,  ..., 0, 0, 0],\n",
              "         [1, 1, 1,  ..., 0, 0, 0],\n",
              "         ...,\n",
              "         [1, 1, 1,  ..., 0, 0, 0],\n",
              "         [1, 1, 1,  ..., 0, 0, 0],\n",
              "         [1, 1, 1,  ..., 0, 0, 0]]),\n",
              " 'input_ids': tensor([[  101, 22226,  2006,  ...,     0,     0,     0],\n",
              "         [  101,  8112,  2000,  ...,     0,     0,     0],\n",
              "         [  101,  5920,  9472,  ...,     0,     0,     0],\n",
              "         ...,\n",
              "         [  101, 19463,  2088,  ...,     0,     0,     0],\n",
              "         [  101, 10079, 14369,  ...,     0,     0,     0],\n",
              "         [  101,  1037,  2450,  ...,     0,     0,     0]]),\n",
              " 'score': tensor([0.8000, 0.4400, 0.5600, 0.7200, 0.8400, 0.0400, 0.9600, 0.6400, 0.4000,\n",
              "         0.6000, 0.6000, 0.4400, 0.5600, 0.0000, 0.2400, 0.8800],\n",
              "        dtype=torch.float64),\n",
              " 'token_type_ids': tensor([[0, 0, 0,  ..., 0, 0, 0],\n",
              "         [0, 0, 0,  ..., 0, 0, 0],\n",
              "         [0, 0, 0,  ..., 0, 0, 0],\n",
              "         ...,\n",
              "         [0, 0, 0,  ..., 0, 0, 0],\n",
              "         [0, 0, 0,  ..., 0, 0, 0],\n",
              "         [0, 0, 0,  ..., 0, 0, 0]])}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 20
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LsvkA72b8Qdc",
        "colab_type": "text"
      },
      "source": [
        "# Model Training And Evaluation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c5dhHU858R_x",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 284,
          "referenced_widgets": [
            "16228d22f42f41c7baae93d2f4f9faa8",
            "5119b17f80854ffebe0e9458eacb778f",
            "aaac04918b01412ab52a647df2ead4d3",
            "2a2abc8d31a64c7e840b0ab9639c7136",
            "08b852d8acc148dcab7586104de73b56",
            "1d0528b5c8ac41e7abd78e6ba51c36b1",
            "174bf646c7ec42caaace3472efd0cce9",
            "8abcd4a7d6b34a38b13678538d945752",
            "a56d60dae6544f75966da9c88832a432",
            "e25554e1d03e44819c0a592cc76a9a40",
            "514a29861a164e05b767d6b10bfc98b5",
            "8f2efda139d74bdfbcefbe7dc9a61d6c",
            "d8ca9f13f86d4957811b157af10551fc",
            "6f294a984012499fa052e068531b51fa",
            "283a878e291a4cb99e4b4d5fe17d4967",
            "046a49a886044c02935ad4d5886cbeed",
            "c815b5730056477b83b7d8ddb23eebac",
            "01d7c17f88b54ff2b0c7babef44191b8",
            "e52809531432458da1ffdf92604dea38",
            "a088d70f081d435ebf6dce0b45165df3",
            "97f4ef1eb5034d1fbe2f8e8a9de0bdb2",
            "861a215eb8d443daa8ca35fa8f57dbf2",
            "8bfdd0e87d864af38e0dc64756304bb1",
            "1bf583ee208547ed8e1f126863f47dba",
            "ecc2382a18cb4b74b2c4f7823383637f",
            "1ef3ada8af3d4f5fa058d0eb7071fe22",
            "dab0d86719e843c68a947b127a470fd6",
            "28f7db0ff5904863a3c94b95368ccc6a",
            "00f3e40d82bc4c128a6254636cdc7548",
            "c96e6b0c61684aff967dc974143ab8c8",
            "14b14b227d8b4f0294c73e7527be743f",
            "8aee81ec8bb145a8b6f126b91d78e317"
          ]
        },
        "outputId": "8029a689-69fb-47ac-a4a2-8988cd9c6542"
      },
      "source": [
        "#initiate model\n",
        "custom_model_cos = ElectraSentenceSimilarityModelCos()\n",
        "custom_model_cos = custom_model_cos.to(device)\n",
        "\n",
        "custom_model_reg = ElectraSentenceSimilarityModelReg()\n",
        "custom_model_reg = custom_model_reg.to(device)\n",
        "\n",
        "custom_model_BertElectra = ElectraBertSentenceSimilarityModelReg()\n",
        "custom_model_BertElectra = custom_model_BertElectra.to(device)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "16228d22f42f41c7baae93d2f4f9faa8",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=467.0, style=ProgressStyle(description_…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "a56d60dae6544f75966da9c88832a432",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=440343552.0, style=ProgressStyle(descri…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Some weights of the model checkpoint at google/electra-base-discriminator were not used when initializing ElectraModel: ['electra.embeddings_project.weight', 'electra.embeddings_project.bias']\n",
            "- This IS expected if you are initializing ElectraModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPretraining model).\n",
            "- This IS NOT expected if you are initializing ElectraModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "c815b5730056477b83b7d8ddb23eebac",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=433.0, style=ProgressStyle(description_…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "ecc2382a18cb4b74b2c4f7823383637f",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=440473133.0, style=ProgressStyle(descri…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1kpeHjm2LgId",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Qw6h03Wy8YoH",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#setup hyper-parameter\n",
        "EPOCHS = 10 # Training epoch\n",
        "#optimizer_cos =AdamW( custom_model_cos.parameters(), lr=1e-4, betas=(0.9, 0.999), eps=1e-6) #optimizer for cosine ouput\n",
        "#optimizer_reg =AdamW(custom_model_reg.parameters(), lr=1e-4,betas=(0.9, 0.999), eps=1e-6) #optimizer for regression output\n",
        "optimizer_BertElec =AdamW(custom_model_BertElectra.parameters(), lr=1e-4,betas=(0.9, 0.999), eps=1e-6) #optimizer for regression output\n",
        "total_steps = len(train_data_loader_cos) * EPOCHS # total number training steps\n",
        "\n",
        "#learning rate scheduler\n",
        "scheduler_cos = get_linear_schedule_with_warmup(  \n",
        "  optimizer_cos,  \n",
        "  num_warmup_steps=0,\n",
        "  num_training_steps=total_steps\n",
        ")  \n",
        "\n",
        "scheduler_reg = get_linear_schedule_with_warmup(  \n",
        "  optimizer_reg,  \n",
        "  num_warmup_steps=0,\n",
        "  num_training_steps=total_steps\n",
        ")  \n",
        "\n",
        "scheduler_BertElec = get_linear_schedule_with_warmup(  \n",
        "  optimizer_BertElec,  \n",
        "  num_warmup_steps=0,\n",
        "  num_training_steps=total_steps\n",
        ")  \n",
        "\n",
        "#MSE loss function\n",
        "loss_fn = nn.MSELoss().to(device)\n",
        "\n",
        "#model type\n",
        "model_type = ['regression', 'cosine']\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kEPsbKGmKXEM",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rv0zo3RFKQ6w",
        "colab_type": "text"
      },
      "source": [
        "####Cosine Similarity Output Model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MUijnI2i8ZwF",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "21e8356d-0bd1-48f4-bc85-47cc956380f9"
      },
      "source": [
        "%%time\n",
        "#Training process start for Cosine similarity output model\n",
        "epochs_score = defaultdict(list) #store scores\n",
        "val_score = 0 #validation score\n",
        "\n",
        "for epoch in range(EPOCHS):\n",
        "  print('Epoch {:}/{:}'.format(epoch + 1, EPOCHS))\n",
        "  print('=='*5)\n",
        "  print()\n",
        "  \n",
        "  #model training initiate\n",
        "  train_loss, train_Pearson_score = model_train(custom_model_cos, train_data_loader_cos, device, loss_fn, optimizer_cos, scheduler_cos, model_type[1])\n",
        "  print('Train loss is: {}, train_Pearson_score is {}'.format(train_loss,train_Pearson_score))\n",
        "  #store values\n",
        "  epochs_score['train_loss'].append(train_loss) #train losss\n",
        "  epochs_score['train_Pearson_score'].append(train_Pearson_score) # pearson correlation\n",
        "\n",
        "\n",
        "  #model validatiaon initiate\n",
        "  val_loss, val_Pearson_score, val_Spearman_score = model_eval(custom_model_cos, dev_data_loader_cos, device, loss_fn, model_type[1])\n",
        "  print('Validation loss is: {}, Validation_Pearson_score is {}'.format(val_loss,val_Pearson_score))\n",
        "  print()\n",
        "  #store values\n",
        "  epochs_score['val_loss'].append(val_loss) #train losss\n",
        "  epochs_score['val_Pearson_score'].append(val_Pearson_score) # pearson correlation\n",
        "  epochs_score['val_Spearman_score'].append(val_Spearman_score) # Spearman correlation\n",
        "\n",
        "  #save model with high validaiton score\n",
        "  if val_Pearson_score >  val_score:\n",
        "    val_score = val_Pearson_score\n",
        "    torch.save(custom_model_cos.state_dict(), 'Electra_cos_best_model.bin')\n",
        "\n",
        "print()\n",
        "print('Training completed......')\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.09950532399945788, train_Pearson_score is 0.16991115754385216\n",
            "Validation loss is: 0.083480271094657, Validation_Pearson_score is 0.5695479550593343\n",
            "\n",
            "Epoch 2/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.0635827765489618, train_Pearson_score is 0.5209959026757578\n",
            "Validation loss is: 0.05338765228999422, Validation_Pearson_score is 0.7019368329109742\n",
            "\n",
            "Epoch 3/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.04079828499298957, train_Pearson_score is 0.7244741614454837\n",
            "Validation loss is: 0.0491655087296633, Validation_Pearson_score is 0.7713373979989414\n",
            "\n",
            "Epoch 4/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.02426913876293434, train_Pearson_score is 0.8466110943335656\n",
            "Validation loss is: 0.039651960530813705, Validation_Pearson_score is 0.7915242242566503\n",
            "\n",
            "Epoch 5/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.015391676352980237, train_Pearson_score is 0.9070612679960413\n",
            "Validation loss is: 0.0388025950957486, Validation_Pearson_score is 0.8010849095814119\n",
            "\n",
            "Epoch 6/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.010622977969857554, train_Pearson_score is 0.936561871594439\n",
            "Validation loss is: 0.03873712533807501, Validation_Pearson_score is 0.7970060385463318\n",
            "\n",
            "Epoch 7/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.007972425611741427, train_Pearson_score is 0.9524319455027801\n",
            "Validation loss is: 0.03728624270792971, Validation_Pearson_score is 0.799817098134002\n",
            "\n",
            "Epoch 8/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.006623854503656427, train_Pearson_score is 0.9601154165725543\n",
            "Validation loss is: 0.03818731017886324, Validation_Pearson_score is 0.7997800475133655\n",
            "\n",
            "Epoch 9/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.005642985812543581, train_Pearson_score is 0.9668631692571796\n",
            "Validation loss is: 0.037846279290920874, Validation_Pearson_score is 0.8008001207336255\n",
            "\n",
            "Epoch 10/10\n",
            "==========\n",
            "\n",
            "Model is training based on cosine ouput!!!\n",
            "Train loss is: 0.005076160492737674, train_Pearson_score is 0.9704664597613241\n",
            "Validation loss is: 0.03803009603251802, Validation_Pearson_score is 0.8031736511254166\n",
            "\n",
            "\n",
            "Training completed......\n",
            "CPU times: user 29min 26s, sys: 17min 16s, total: 46min 42s\n",
            "Wall time: 46min 48s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9U7kDLmgijkT",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#clear unused GPU\n",
        "with torch.no_grad():\n",
        "  torch.cuda.empty_cache()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qwUCkEZR9bTU",
        "colab_type": "text"
      },
      "source": [
        "####Plot Model Learning Curve"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "10AYfmaw8ht1",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 410
        },
        "outputId": "936530b5-9e49-4de4-f7ec-df5edd72ed88"
      },
      "source": [
        "#plot training and validation curve \n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns\n",
        "train_score = epochs_score['train_Pearson_score'] # train_score\n",
        "valid_score = epochs_score['val_Pearson_score'] # valid_score\n",
        " \n",
        "\n",
        "sns.set(style='darkgrid') #set plot style\n",
        "sns.set(rc={'figure.figsize':(9,6)}) # set figure size\n",
        "plt.plot(train_score ,'o-b', label= 'Train_score' )#plot train score\n",
        "plt.plot(valid_score, '^-r', label='Valid_score') # plot valid score\n",
        "plt.xticks(range(0, EPOCHS, 1)) #set xlabel location\n",
        "plt.title('Model Training Curve(Siamese Network)') #  title\n",
        "plt.ylabel('Pearson Coefficient') # ylabel\n",
        "plt.xlabel('No. of epochs') # xlabel\n",
        "plt.ylim([0,1]) #set the ylabel limit\n",
        "plt.legend(loc=4) #set label location\n",
        "plt.show()\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 648x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gFT0iJLvMF_2",
        "colab_type": "text"
      },
      "source": [
        "####Evaluation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "r_vPbfRCfH_l",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "9af87264-047f-40de-b743-2a266dd6108c"
      },
      "source": [
        "#evaluate on test data cosine similarity output\n",
        "test_loss, test_Pearson_score, test_Spearman_score = model_eval(custom_model_cos, \n",
        "                                                                  test_data_loader_cos, \n",
        "                                                                  device, \n",
        "                                                                  loss_fn,\n",
        "                                                                  model_type[1])\n",
        "print(test_loss, test_Pearson_score, test_Spearman_score)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.04180855269078165 0.7810846984909834 0.7723773028372443\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "60by7AVaHxfp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "f4f2810c-f593-489f-f301-d3381aae96c2"
      },
      "source": [
        "#dev dataset\n",
        "dev_loss, dev_Pearson_score, dev_Spearman_score = model_eval(custom_model_cos, \n",
        "                                                              dev_data_loader_cos, \n",
        "                                                              device, \n",
        "                                                              loss_fn,\n",
        "                                                              model_type[1])\n",
        "print(dev_loss, dev_Pearson_score, dev_Spearman_score)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.03797970866744823 0.805442073895768 0.7950363803486822\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2vzfTygis796",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6HAbCOb0J7qK",
        "colab_type": "text"
      },
      "source": [
        "###Regression Output model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FflGTmV3fH4w",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "bff7f75c-1072-43de-de26-0bda028ae052"
      },
      "source": [
        "%%time\n",
        "#Training process start for regression output model\n",
        "epochs_score_reg = defaultdict(list) #store score from each epochs\n",
        "val_score_reg = 0 # validation score\n",
        "\n",
        "for epoch in range(EPOCHS):\n",
        "  print('Epoch {:}/{:}'.format(epoch + 1, EPOCHS))\n",
        "  print('=='*5)\n",
        "  print()\n",
        "  \n",
        "  #model training initiate\n",
        "  train_loss, train_Pearson_score = model_train(custom_model_reg, train_data_loader_reg, device, loss_fn, optimizer_reg, scheduler_reg, model_type[0])\n",
        "  print('Train loss is: {}, train_Pearson_score is {}'.format(train_loss,train_Pearson_score))\n",
        "  #store values\n",
        "  epochs_score_reg['train_loss'].append(train_loss) #train losss\n",
        "  epochs_score_reg['train_Pearson_score'].append(train_Pearson_score) # pearson correlation\n",
        "\n",
        "\n",
        "  #model validatiaon initiate\n",
        "  val_loss, val_Pearson_score, val_Spearman_score = model_eval(custom_model_reg, dev_data_loader_reg, device, loss_fn, model_type[0])\n",
        "  print('Validation loss is: {}, Validation_Pearson_score is {}'.format(val_loss,val_Pearson_score))\n",
        "  print()\n",
        "  #store values\n",
        "  epochs_score_reg['val_loss'].append(val_loss) #train losss\n",
        "  epochs_score_reg['val_Pearson_score'].append(val_Pearson_score) # pearson correlation\n",
        "  epochs_score_reg['val_Spearman_score'].append(val_Spearman_score) # Spearman correlation\n",
        "\n",
        "  #save model with high validaiton score\n",
        "  if val_Pearson_score >  val_score_reg:\n",
        "    val_score_reg = val_Pearson_score\n",
        "    torch.save(custom_model_reg.state_dict(), 'Electra_reg_best_model.bin')\n",
        "\n",
        "print()\n",
        "print('Training completed......')\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.050652560814180306, train_Pearson_score is 0.6891675863623151\n",
            "Validation loss is: 0.021922666678561808, Validation_Pearson_score is 0.8917225831912159\n",
            "\n",
            "Epoch 2/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.018781975511875416, train_Pearson_score is 0.8946080221217818\n",
            "Validation loss is: 0.01685201697685617, Validation_Pearson_score is 0.9034971004409484\n",
            "\n",
            "Epoch 3/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.011540286561163763, train_Pearson_score is 0.9379049437114456\n",
            "Validation loss is: 0.016722207651176352, Validation_Pearson_score is 0.9120235776823128\n",
            "\n",
            "Epoch 4/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.007917642034590245, train_Pearson_score is 0.9563160065294027\n",
            "Validation loss is: 0.017094035930138952, Validation_Pearson_score is 0.9092364647164214\n",
            "\n",
            "Epoch 5/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.005840677351483868, train_Pearson_score is 0.9679885931874929\n",
            "Validation loss is: 0.015190317335122444, Validation_Pearson_score is 0.9107435146767009\n",
            "\n",
            "Epoch 6/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.004417946743261483, train_Pearson_score is 0.9752632271652639\n",
            "Validation loss is: 0.018041634129637732, Validation_Pearson_score is 0.9122533926052852\n",
            "\n",
            "Epoch 7/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.003647627816018131, train_Pearson_score is 0.979409361462817\n",
            "Validation loss is: 0.016129988552804325, Validation_Pearson_score is 0.9143041430895628\n",
            "\n",
            "Epoch 8/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.00318214518419053, train_Pearson_score is 0.9825782392883926\n",
            "Validation loss is: 0.015059206099744806, Validation_Pearson_score is 0.9138197381537501\n",
            "\n",
            "Epoch 9/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.0026551624656551416, train_Pearson_score is 0.9847204205746333\n",
            "Validation loss is: 0.01646571441930025, Validation_Pearson_score is 0.9122034288035596\n",
            "\n",
            "Epoch 10/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.002396400293542279, train_Pearson_score is 0.9858737276451797\n",
            "Validation loss is: 0.01598322521300709, Validation_Pearson_score is 0.9119348372334096\n",
            "\n",
            "\n",
            "Training completed......\n",
            "CPU times: user 24min 2s, sys: 14min 32s, total: 38min 34s\n",
            "Wall time: 38min 39s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BaDuyqKnnIZb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#clear unused GPU\n",
        "with torch.no_grad():\n",
        "  torch.cuda.empty_cache()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xEAG5aZyg1Pb",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 410
        },
        "outputId": "36ce913b-32be-4bfc-e324-4828f138cf60"
      },
      "source": [
        "#plot training and validation curve for Regression\n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns\n",
        "train_score_reg = epochs_score_reg['train_Pearson_score'] # train_score\n",
        "valid_score_reg = epochs_score_reg['val_Pearson_score'] # valid_score\n",
        " \n",
        "\n",
        "sns.set(style='darkgrid') #set plot style\n",
        "sns.set(rc={'figure.figsize':(9,6)}) # set figure size\n",
        "plt.plot(train_score_reg ,'o-b', label= 'Train_score' )#plot train score\n",
        "plt.plot(valid_score_reg, '*-r', label='Valid_score') # plot valid score\n",
        "plt.xticks(range(0, 10, 1)) #set xlabel location\n",
        "plt.title('Model Training Curve(Regression Output)') #  title\n",
        "plt.ylabel('Pearson Coefficient') # ylabel\n",
        "plt.xlabel('No. of epochs') # xlabel\n",
        "plt.ylim([0.6,1]) #set the ylabel limit\n",
        "plt.legend(loc=4) #set label location\n",
        "plt.show()\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 648x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k6BS7y1_4tdP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "s-O1Fy6MMU-y",
        "colab_type": "text"
      },
      "source": [
        "#### Evaluation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tIIrRO1H8qC2",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "9b96a22f-9a4e-42ef-9d3b-ead603075928"
      },
      "source": [
        "#evaluate on test data \n",
        "test_loss_reg, test_Pearson_score_reg, test_Spearman_score_reg = model_eval(custom_model_reg, \n",
        "                                                                            test_data_loader_reg, \n",
        "                                                                            device, \n",
        "                                                                            loss_fn,\n",
        "                                                                            model_type[0])\n",
        "print(test_loss_reg, test_Pearson_score_reg, test_Spearman_score_reg)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.01859163236275146 0.9011378373625223 0.8859287003278358\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KSbw9lCM5ukP",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "5f4be20c-1e72-43c9-f31a-60b7dffea655"
      },
      "source": [
        "#evaluate on dev data \n",
        "dev_loss_reg, dev_Pearson_score_reg, dev_Spearman_score_reg = model_eval(custom_model_reg, \n",
        "                                                                        dev_data_loader_reg, \n",
        "                                                                        device, \n",
        "                                                                        loss_fn,\n",
        "                                                                        model_type[0])\n",
        "print(dev_loss_reg, dev_Pearson_score_reg, dev_Spearman_score_reg)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.01598502778785026 0.912845530614849 0.897061857003664\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MxQHeAoHGKeq",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KPyEnm-zGK05",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "792df5be-bf27-4675-b029-9424f24cde04"
      },
      "source": [
        "%%time\n",
        "#Training process start for regression output model\n",
        "epochs_score_bertelctra = defaultdict(list) #store score from each epochs\n",
        "val_score_bertelctra = 0 # validation score\n",
        "\n",
        "for epoch in range(EPOCHS):\n",
        "  print('Epoch {:}/{:}'.format(epoch + 1, EPOCHS))\n",
        "  print('=='*5)\n",
        "  print()\n",
        "  \n",
        "  #model training initiate\n",
        "  train_loss_be, train_Pearson_score_be = model_train(custom_model_BertElectra, train_data_loader_reg, device, loss_fn, optimizer_BertElec, scheduler_BertElec, model_type[0])\n",
        "  print('Train loss is: {}, train_Pearson_score is {}'.format(train_loss_be,train_Pearson_score_be))\n",
        "  #store values\n",
        "  epochs_score_bertelctra['train_loss'].append(train_loss_be) #train losss\n",
        "  epochs_score_bertelctra['train_Pearson_score'].append(train_Pearson_score_be) # pearson correlation\n",
        "\n",
        "\n",
        "  #model validatiaon initiate\n",
        "  val_loss_be, val_Pearson_score_be, val_Spearman_score_be = model_eval(custom_model_BertElectra, dev_data_loader_reg, device, loss_fn, model_type[0])\n",
        "  print('Validation loss is: {}, Validation_Pearson_score is {}'.format(val_loss_be, val_Pearson_score_be))\n",
        "  print()\n",
        "  #store values\n",
        "  epochs_score_bertelctra['val_loss'].append(val_loss_be) #train losss\n",
        "  epochs_score_bertelctra['val_Pearson_score'].append(val_Pearson_score_be) # pearson correlation\n",
        "  epochs_score_bertelctra['val_Spearman_score'].append(val_Spearman_score_be) # Spearman correlation\n",
        "\n",
        "  #save model with high validaiton score\n",
        "  if val_Pearson_score_be >  val_score_bertelctra:\n",
        "    val_score_bertelctra = val_Pearson_score_be\n",
        "    torch.save(custom_model_BertElectra.state_dict(), 'BertElectra_reg_best_model.bin')\n",
        "\n",
        "print()\n",
        "print('Training completed......')\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.053620093846176234, train_Pearson_score is 0.7373595120184019\n",
            "Validation loss is: 0.027191743888753526, Validation_Pearson_score is 0.8635254272313446\n",
            "\n",
            "Epoch 2/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.022851546056982544, train_Pearson_score is 0.8729085999626687\n",
            "Validation loss is: 0.022707672403293088, Validation_Pearson_score is 0.8817603416858175\n",
            "\n",
            "Epoch 3/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.01334459764669494, train_Pearson_score is 0.9317394488303402\n",
            "Validation loss is: 0.024423532088861822, Validation_Pearson_score is 0.8783918995089054\n",
            "\n",
            "Epoch 4/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.007975044652509193, train_Pearson_score is 0.9566414857927422\n",
            "Validation loss is: 0.019007798730138133, Validation_Pearson_score is 0.887763307215896\n",
            "\n",
            "Epoch 5/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.005380660124097226, train_Pearson_score is 0.9698714032269854\n",
            "Validation loss is: 0.01824465973619768, Validation_Pearson_score is 0.8922311341425259\n",
            "\n",
            "Epoch 6/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.003908570985367987, train_Pearson_score is 0.9793484324662642\n",
            "Validation loss is: 0.01775022907222205, Validation_Pearson_score is 0.8890292911881658\n",
            "\n",
            "Epoch 7/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.002674137138334724, train_Pearson_score is 0.9851006835277164\n",
            "Validation loss is: 0.01747500644303224, Validation_Pearson_score is 0.8929817238789869\n",
            "\n",
            "Epoch 8/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.0020535807956346412, train_Pearson_score is 0.988137947227757\n",
            "Validation loss is: 0.01789540118300059, Validation_Pearson_score is 0.8984364939532374\n",
            "\n",
            "Epoch 9/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.0016436290951484503, train_Pearson_score is 0.9906159669806486\n",
            "Validation loss is: 0.017641087196410655, Validation_Pearson_score is 0.8944985558096498\n",
            "\n",
            "Epoch 10/10\n",
            "==========\n",
            "\n",
            "Model is training based on regression ouput!!!\n",
            "Train loss is: 0.0014277223996436482, train_Pearson_score is 0.9917276000268873\n",
            "Validation loss is: 0.017741961914927086, Validation_Pearson_score is 0.8951705492031675\n",
            "\n",
            "\n",
            "Training completed......\n",
            "CPU times: user 29min 42s, sys: 18min 17s, total: 47min 59s\n",
            "Wall time: 48min 15s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lQoPLi6GGKYR",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 410
        },
        "outputId": "6a3f65ee-d2d3-411e-c38e-c9338b49712e"
      },
      "source": [
        "#plot training and validation curve for Bert+Electra combined model\n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns\n",
        "train_score_be = epochs_score_bertelctra['train_Pearson_score'] # train_score\n",
        "valid_score_be = epochs_score_bertelctra['val_Pearson_score'] # valid_score\n",
        " \n",
        "\n",
        "sns.set(style='darkgrid') #set plot style\n",
        "sns.set(rc={'figure.figsize':(9,6)}) # set figure size\n",
        "plt.plot(train_score_be ,'o-b', label= 'Train_score' )#plot train score\n",
        "plt.plot(valid_score_be, '*-r', label='Valid_score') # plot valid score\n",
        "plt.xticks(range(0, 10, 1)) #set xlabel location\n",
        "plt.title('Model Training Curve(Bert+Electra)') #  title\n",
        "plt.ylabel('Pearson Coefficient') # ylabel\n",
        "plt.xlabel('No. of epochs') # xlabel\n",
        "plt.ylim([0.6,1]) #set the ylabel limit\n",
        "plt.legend(loc=4) #set label location\n",
        "plt.show()\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 648x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Cz0H3o0LHRIB",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "35508408-e06c-4c91-9dee-5bb1a2dfa4b3"
      },
      "source": [
        "#evaluate on test data \n",
        "test_loss_be, test_Pearson_score_be, test_Spearman_score_be = model_eval(custom_model_BertElectra, \n",
        "                                                                        test_data_loader_reg, \n",
        "                                                                        device, \n",
        "                                                                        loss_fn,\n",
        "                                                                        model_type[0])\n",
        "print(test_loss_be, test_Pearson_score_be, test_Spearman_score_be)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.024075748249032032 0.8549361319865953 0.8238797900145182\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KZnAlBQwyYvS",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "2578d569-6df2-477a-cef4-b1811514fc07"
      },
      "source": [
        "test_loss_be, test_Pearson_score_be, test_Spearman_score_be = model_eval(custom_model_BertElectra, \n",
        "                                                                        dev_data_loader_reg, \n",
        "                                                                        device, \n",
        "                                                                        loss_fn,\n",
        "                                                                        model_type[0])\n",
        "print(test_loss_be, test_Pearson_score_be, test_Spearman_score_be)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "0.017782565087397048 0.8923963216053599 0.8734975231814652\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fZFPk7VfHRBo",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "srhjKxXIJiaF",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 124
        },
        "outputId": "af68edca-190a-4471-b3ec-5555695504b7"
      },
      "source": [
        "#show in pandas dataframe dev_Spearman_score\n",
        "electra_df = pd.DataFrame( data=[['Electra-CosOutput', 0.805442073895768, 0.7810846984909834, 'Pearson',],\n",
        "                    ['Electra-CosOutput',  0.7950363803486822,  0.7723773028372443, 'Spearman'],\n",
        "                    ['Electra-RegOutput', 0.912845530614849, 0.9011378373625223, 'Pearson'],\n",
        "                    ['Electra-RegOutput', 0.897061857003664, 0.8859287003278358, 'Spearman']\n",
        "                    ], \n",
        "             columns=['Method-Types', 'DevSet', 'TestSet', 'Metric'])\n",
        "\n",
        "#highlight the maximum in a column with yellow.\n",
        "def highlight_max(val):\n",
        "  \n",
        "    is_max = val == val.max()\n",
        "    return ['background-color: yellow' if v else '' for v in is_max]\n",
        "\n",
        "electra_df.style.apply(highlight_max, subset=['DevSet', 'TestSet'])\n",
        "\n",
        "\n",
        "#import plotly.figure_factory as ff \n",
        "#ff.create_table(electra_df)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<style  type=\"text/css\" >\n",
              "    #T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row2_col1 {\n",
              "            background-color:  yellow;\n",
              "        }    #T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row2_col2 {\n",
              "            background-color:  yellow;\n",
              "        }</style><table id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002\" ><thead>    <tr>        <th class=\"blank level0\" ></th>        <th class=\"col_heading level0 col0\" >Method-Types</th>        <th class=\"col_heading level0 col1\" >DevSet</th>        <th class=\"col_heading level0 col2\" >TestSet</th>        <th class=\"col_heading level0 col3\" >Metric</th>    </tr></thead><tbody>\n",
              "                <tr>\n",
              "                        <th id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002level0_row0\" class=\"row_heading level0 row0\" >0</th>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row0_col0\" class=\"data row0 col0\" >Electra-CosOutput</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row0_col1\" class=\"data row0 col1\" >0.805442</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row0_col2\" class=\"data row0 col2\" >0.781085</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row0_col3\" class=\"data row0 col3\" >Pearson</td>\n",
              "            </tr>\n",
              "            <tr>\n",
              "                        <th id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002level0_row1\" class=\"row_heading level0 row1\" >1</th>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row1_col0\" class=\"data row1 col0\" >Electra-CosOutput</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row1_col1\" class=\"data row1 col1\" >0.795036</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row1_col2\" class=\"data row1 col2\" >0.772377</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row1_col3\" class=\"data row1 col3\" >Spearman</td>\n",
              "            </tr>\n",
              "            <tr>\n",
              "                        <th id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002level0_row2\" class=\"row_heading level0 row2\" >2</th>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row2_col0\" class=\"data row2 col0\" >Electra-RegOutput</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row2_col1\" class=\"data row2 col1\" >0.912846</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row2_col2\" class=\"data row2 col2\" >0.901138</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row2_col3\" class=\"data row2 col3\" >Pearson</td>\n",
              "            </tr>\n",
              "            <tr>\n",
              "                        <th id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002level0_row3\" class=\"row_heading level0 row3\" >3</th>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row3_col0\" class=\"data row3 col0\" >Electra-RegOutput</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row3_col1\" class=\"data row3 col1\" >0.897062</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row3_col2\" class=\"data row3 col2\" >0.885929</td>\n",
              "                        <td id=\"T_b0cacaa6_d17d_11ea_8e02_0242ac1c0002row3_col3\" class=\"data row3 col3\" >Spearman</td>\n",
              "            </tr>\n",
              "    </tbody></table>"
            ],
            "text/plain": [
              "<pandas.io.formats.style.Styler at 0x7f906ab18ac8>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 38
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Z2h2dtz1O0LM",
        "colab_type": "text"
      },
      "source": [
        "####Observation:\n",
        "Electra-RegOuput i.e one regression layer on top of pretained-ELECTRA model gives higest result in both dataset(dev and test)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rXoAnRSjOy_b",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "05tRdGLV9Ekl",
        "colab_type": "text"
      },
      "source": [
        "# Save and Load model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5uAlwxqe9FuM",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#save model\n",
        "#torch.save(custom_model.state_dict(), '/content/drive/My Drive/Google_Colab/Electra_best_model.bin')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1K7gOPwO9WBJ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 105
        },
        "outputId": "884151d7-ad7a-4815-8a70-f16feabf31a0"
      },
      "source": [
        "#load model and evaluate on test data\n",
        "fine_tuned_model = ElectraSentenceSimilarityModelCos()\n",
        "fine_tuned_model.load_state_dict(torch.load('/content/best_modElectra_best_model.bin'))\n",
        "test_loss, test_Pearson_score = model_eval(\n",
        "                                      fine_tuned_model.to(device),\n",
        "                                      test_data_loader, \n",
        "                                      device, \n",
        "                                      loss_fn)\n",
        "print(test_loss, test_Pearson_score)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Some weights of the model checkpoint at google/electra-base-discriminator were not used when initializing ElectraModel: ['electra.embeddings_project.weight', 'electra.embeddings_project.bias']\n",
            "- This IS expected if you are initializing ElectraModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPretraining model).\n",
            "- This IS NOT expected if you are initializing ElectraModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "0.04113160383127544 0.7887276552132803\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DHn2AAjryE3y",
        "colab_type": "text"
      },
      "source": [
        "# References:\n",
        "1. Chris McCormick and Nick Ryan. (2019, May 14). BERT Word Embeddings Tutorial. Retrieved from http://www.mccormickml.com , https://mccormickml.com/2019/05/14/BERT-word-embeddings-tutorial/\n",
        "\n",
        "2. [Sentiment Analysis with BERT](https://colab.research.google.com/drive/1PHv-IRLPCtv7oTcIGbsgZHqrB5LPvB7S#scrollTo=1zhHoFNsxufs&forceEdit=true&sandboxMode=true)\n",
        "\n",
        "3. [Hugging Face Transformer](https://huggingface.co/transformers/)\n",
        "\n",
        "4. [The Illustrated BERT, ELMo, and co. (How NLP Cracked Transfer Learning)](https://jalammar.github.io/illustrated-bert/)\n",
        "\n",
        "5. [ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators](https://arxiv.org/abs/2003.10555)\n"
      ]
    }
  ]
}