cifar10_smallVGG.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "smallVGG",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QEj6V2piUnFZ",
        "colab_type": "text"
      },
      "source": [
        "![alt text](https://i.imgur.com/0cvnBXi.png)\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J7y-5g7qPrMN",
        "colab_type": "code",
        "outputId": "240cad41-10ce-47a5-f38a-f706a4ba2a52",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "%tensorflow_version 1.x \n",
        "\n",
        "#Train a simple deep CNN on the CIFAR10 small images dataset\n",
        "from __future__ import print_function\n",
        "import keras\n",
        "from keras.datasets import cifar10\n",
        "from keras.preprocessing.image import ImageDataGenerator\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Dense, Dropout, Activation, Flatten\n",
        "from keras.layers import Conv2D, MaxPooling2D, BatchNormalization, ZeroPadding2D\n",
        "import os\n",
        "\n",
        "batch_size = 128\n",
        "num_classes = 10\n",
        "epochs = 30\n",
        "data_augmentation = True\n",
        "num_predictions = 20\n",
        "\n",
        "\n",
        "# The data, split between train and test sets:\n",
        "(x_train, y_train), (x_test, y_test) = cifar10.load_data()\n",
        "print('x_train shape:', x_train.shape)\n",
        "print(x_train.shape[0], 'train samples')\n",
        "print(x_test.shape[0], 'test samples')\n",
        "\n",
        "# Convert class vectors to binary class matrices.\n",
        "y_train = keras.utils.to_categorical(y_train, num_classes)\n",
        "y_test = keras.utils.to_categorical(y_test, num_classes)\n",
        "\n",
        "model = Sequential()\n",
        "model.add(Conv2D(32, (3, 3), padding='same',\n",
        "                 input_shape=x_train.shape[1:]))\n",
        "model.add(Activation('relu'))\n",
        "model.add(Conv2D(32, (3, 3)))\n",
        "model.add(Activation('relu'))\n",
        "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "model.add(Dropout(0.25))\n",
        "\n",
        "model.add(Conv2D(64, (3, 3), padding='same'))\n",
        "model.add(Activation('relu'))\n",
        "model.add(Conv2D(64, (3, 3)))\n",
        "model.add(Activation('relu'))\n",
        "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "model.add(Dropout(0.25))\n",
        "\n",
        "model.add(Conv2D(128, (3, 3), padding='same'))\n",
        "model.add(Activation('relu'))\n",
        "model.add(Conv2D(128, (3, 3)))\n",
        "model.add(Activation('relu'))\n",
        "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "model.add(Dropout(0.25))\n",
        "\n",
        "model.add(Conv2D(512, (3, 3), padding='same'))\n",
        "model.add(Activation('relu'))\n",
        "model.add(Dropout(0.5))\n",
        "model.add(Conv2D(512, (1, 1), padding='same'))\n",
        "model.add(Activation('relu'))\n",
        "model.add(Dropout(0.5))\n",
        "\n",
        "model.add(Conv2D(num_classes, (1, 1), padding='same'))\n",
        "model.add(Flatten())\n",
        "model.add(Dense(num_classes))\n",
        "model.add(Activation('softmax'))\n",
        "\n",
        "x_train = x_train.astype('float32')\n",
        "x_test = x_test.astype('float32')\n",
        "x_train /= 255\n",
        "x_test /= 255\n",
        "\n",
        "if not data_augmentation:\n",
        "    print('Not using data augmentation.')\n",
        "    model.fit(x_train, y_train,\n",
        "              batch_size=batch_size,\n",
        "              epochs=epochs,\n",
        "              validation_data=(x_test, y_test),\n",
        "              shuffle=True)\n",
        "else:\n",
        "    print('Using real-time data augmentation.')\n",
        "    # This will do preprocessing and realtime data augmentation:\n",
        "    datagen = ImageDataGenerator(\n",
        "        featurewise_center=False,  # set input mean to 0 over the dataset\n",
        "        samplewise_center=False,  # set each sample mean to 0\n",
        "        featurewise_std_normalization=False,  # divide inputs by std of the dataset\n",
        "        samplewise_std_normalization=False,  # divide each input by its std\n",
        "        zca_whitening=False,  # apply ZCA whitening\n",
        "        zca_epsilon=1e-06,  # epsilon for ZCA whitening\n",
        "        rotation_range=0,  # randomly rotate images in the range (degrees, 0 to 180)\n",
        "        # randomly shift images horizontally (fraction of total width)\n",
        "        width_shift_range=0.1,\n",
        "        # randomly shift images vertically (fraction of total height)\n",
        "        height_shift_range=0.1,\n",
        "        shear_range=0.,  # set range for random shear\n",
        "        zoom_range=0.,  # set range for random zoom\n",
        "        channel_shift_range=0.,  # set range for random channel shifts\n",
        "        # set mode for filling points outside the input boundaries\n",
        "        fill_mode='nearest',\n",
        "        cval=0.,  # value used for fill_mode = \"constant\"\n",
        "        horizontal_flip=True,  # randomly flip images\n",
        "        vertical_flip=False,  # randomly flip images\n",
        "        # set rescaling factor (applied before any other transformation)\n",
        "        rescale=None,\n",
        "        # set function that will be applied on each input\n",
        "        preprocessing_function=None,\n",
        "        # image data format, either \"channels_first\" or \"channels_last\"\n",
        "        data_format=None,\n",
        "        # fraction of images reserved for validation (strictly between 0 and 1)\n",
        "        validation_split=0.0)\n",
        "\n",
        "    # Compute quantities required for feature-wise normalization\n",
        "    # (std, mean, and principal components if ZCA whitening is applied).\n",
        "    datagen.fit(x_train)\n",
        "\n",
        "    \n",
        "    # initiate optimizers\n",
        "\n",
        "    adam_opt = keras.optimizers.Adam(learning_rate=0.001, beta_1=0.9, beta_2=0.999, amsgrad=False)\n",
        "    ams_opt = keras.optimizers.Adam(learning_rate=0.001, beta_1=0.9, beta_2=0.999, amsgrad=False)\n",
        "\n",
        "    # Let's train the model\n",
        "    model.compile(loss='categorical_crossentropy',\n",
        "                  optimizer=adam_opt,\n",
        "                  metrics=['accuracy'])\n",
        "    model.save_weights('model.h5')\n",
        "\n",
        "    # Fit the model on the batches generated by datagen.flow().\n",
        "    history_adam = model.fit_generator(datagen.flow(x_train, y_train,\n",
        "                                     batch_size=batch_size),\n",
        "                        epochs=epochs,\n",
        "                        validation_data=(x_test, y_test),\n",
        "                        workers=4)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "TensorFlow 1.x selected.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz\n",
            "170500096/170498071 [==============================] - 2s 0us/step\n",
            "x_train shape: (50000, 32, 32, 3)\n",
            "50000 train samples\n",
            "10000 test samples\n",
            "WARNING:tensorflow:From /tensorflow-1.15.2/python3.6/tensorflow_core/python/ops/resource_variable_ops.py:1630: calling BaseResourceVariable.__init__ (from tensorflow.python.ops.resource_variable_ops) with constraint is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "If using Keras pass *_constraint arguments to layers.\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:4070: The name tf.nn.max_pool is deprecated. Please use tf.nn.max_pool2d instead.\n",
            "\n",
            "Using real-time data augmentation.\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:422: The name tf.global_variables is deprecated. Please use tf.compat.v1.global_variables instead.\n",
            "\n",
            "Epoch 1/30\n",
            "391/391 [==============================] - 407s 1s/step - loss: 1.8787 - accuracy: 0.2727 - val_loss: 1.4578 - val_accuracy: 0.4557\n",
            "Epoch 2/30\n",
            "391/391 [==============================] - 406s 1s/step - loss: 1.4546 - accuracy: 0.4634 - val_loss: 1.2000 - val_accuracy: 0.5637\n",
            "Epoch 3/30\n",
            "391/391 [==============================] - 403s 1s/step - loss: 1.2519 - accuracy: 0.5504 - val_loss: 1.0386 - val_accuracy: 0.6276\n",
            "Epoch 4/30\n",
            "391/391 [==============================] - 403s 1s/step - loss: 1.1214 - accuracy: 0.5999 - val_loss: 0.9321 - val_accuracy: 0.6685\n",
            "Epoch 5/30\n",
            "391/391 [==============================] - 410s 1s/step - loss: 1.0265 - accuracy: 0.6394 - val_loss: 0.8683 - val_accuracy: 0.6936\n",
            "Epoch 6/30\n",
            "391/391 [==============================] - 397s 1s/step - loss: 0.9744 - accuracy: 0.6589 - val_loss: 0.8672 - val_accuracy: 0.7016\n",
            "Epoch 7/30\n",
            "391/391 [==============================] - 393s 1s/step - loss: 0.9174 - accuracy: 0.6810 - val_loss: 0.9028 - val_accuracy: 0.6891\n",
            "Epoch 8/30\n",
            "391/391 [==============================] - 392s 1s/step - loss: 0.8928 - accuracy: 0.6896 - val_loss: 0.7295 - val_accuracy: 0.7505\n",
            "Epoch 9/30\n",
            "391/391 [==============================] - 387s 989ms/step - loss: 0.8587 - accuracy: 0.7028 - val_loss: 0.7983 - val_accuracy: 0.7282\n",
            "Epoch 10/30\n",
            "391/391 [==============================] - 390s 997ms/step - loss: 0.8329 - accuracy: 0.7097 - val_loss: 0.7609 - val_accuracy: 0.7381\n",
            "Epoch 11/30\n",
            "391/391 [==============================] - 390s 998ms/step - loss: 0.8042 - accuracy: 0.7208 - val_loss: 0.7413 - val_accuracy: 0.7473\n",
            "Epoch 12/30\n",
            "391/391 [==============================] - 385s 985ms/step - loss: 0.7865 - accuracy: 0.7297 - val_loss: 0.6838 - val_accuracy: 0.7661\n",
            "Epoch 13/30\n",
            "391/391 [==============================] - 388s 993ms/step - loss: 0.7713 - accuracy: 0.7338 - val_loss: 0.7169 - val_accuracy: 0.7568\n",
            "Epoch 14/30\n",
            "391/391 [==============================] - 388s 993ms/step - loss: 0.7517 - accuracy: 0.7386 - val_loss: 0.6892 - val_accuracy: 0.7644\n",
            "Epoch 15/30\n",
            "391/391 [==============================] - 392s 1s/step - loss: 0.7501 - accuracy: 0.7419 - val_loss: 0.6531 - val_accuracy: 0.7801\n",
            "Epoch 16/30\n",
            "391/391 [==============================] - 392s 1s/step - loss: 0.7336 - accuracy: 0.7473 - val_loss: 0.6712 - val_accuracy: 0.7690\n",
            "Epoch 17/30\n",
            "391/391 [==============================] - 389s 994ms/step - loss: 0.7209 - accuracy: 0.7532 - val_loss: 0.6273 - val_accuracy: 0.7850\n",
            "Epoch 18/30\n",
            "391/391 [==============================] - 393s 1s/step - loss: 0.7027 - accuracy: 0.7612 - val_loss: 0.6236 - val_accuracy: 0.7832\n",
            "Epoch 19/30\n",
            "391/391 [==============================] - 392s 1s/step - loss: 0.6936 - accuracy: 0.7621 - val_loss: 0.6192 - val_accuracy: 0.7902\n",
            "Epoch 20/30\n",
            "391/391 [==============================] - 385s 983ms/step - loss: 0.6892 - accuracy: 0.7628 - val_loss: 0.6754 - val_accuracy: 0.7698\n",
            "Epoch 21/30\n",
            "391/391 [==============================] - 389s 994ms/step - loss: 0.6877 - accuracy: 0.7647 - val_loss: 0.5927 - val_accuracy: 0.7969\n",
            "Epoch 22/30\n",
            "391/391 [==============================] - 385s 985ms/step - loss: 0.6718 - accuracy: 0.7689 - val_loss: 0.5988 - val_accuracy: 0.7987\n",
            "Epoch 23/30\n",
            "391/391 [==============================] - 386s 987ms/step - loss: 0.6684 - accuracy: 0.7701 - val_loss: 0.5906 - val_accuracy: 0.7996\n",
            "Epoch 24/30\n",
            "391/391 [==============================] - 394s 1s/step - loss: 0.6627 - accuracy: 0.7747 - val_loss: 0.6473 - val_accuracy: 0.7839\n",
            "Epoch 25/30\n",
            "391/391 [==============================] - 385s 984ms/step - loss: 0.6581 - accuracy: 0.7734 - val_loss: 0.6284 - val_accuracy: 0.7869\n",
            "Epoch 26/30\n",
            "391/391 [==============================] - 387s 989ms/step - loss: 0.6511 - accuracy: 0.7762 - val_loss: 0.5933 - val_accuracy: 0.8003\n",
            "Epoch 27/30\n",
            "391/391 [==============================] - 383s 980ms/step - loss: 0.6448 - accuracy: 0.7804 - val_loss: 0.5976 - val_accuracy: 0.7937\n",
            "Epoch 28/30\n",
            "391/391 [==============================] - 381s 975ms/step - loss: 0.6382 - accuracy: 0.7830 - val_loss: 0.5926 - val_accuracy: 0.7984\n",
            "Epoch 29/30\n",
            "391/391 [==============================] - 383s 980ms/step - loss: 0.6302 - accuracy: 0.7846 - val_loss: 0.5555 - val_accuracy: 0.8107\n",
            "Epoch 30/30\n",
            "391/391 [==============================] - 382s 977ms/step - loss: 0.6274 - accuracy: 0.7869 - val_loss: 0.6061 - val_accuracy: 0.7945\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Fj6ln3XFkGNO",
        "colab_type": "code",
        "outputId": "fb639872-f324-464e-a4e3-5135221fe09f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "    model.load_weights('model.h5')\n",
        "    model.compile(loss='categorical_crossentropy',\n",
        "                  optimizer=ams_opt,\n",
        "                  metrics=['accuracy'])\n",
        "    history_ams = model.fit_generator(datagen.flow(x_train, y_train,\n",
        "                                     batch_size=batch_size),\n",
        "                        epochs=epochs,\n",
        "                        validation_data=(x_test, y_test),\n",
        "                        workers=4)\n",
        "\n",
        "# Score trained model.\n",
        "scores = model.evaluate(x_test, y_test, verbose=1)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/30\n",
            "391/391 [==============================] - 383s 980ms/step - loss: 1.8102 - accuracy: 0.3095 - val_loss: 1.4680 - val_accuracy: 0.4492\n",
            "Epoch 2/30\n",
            "391/391 [==============================] - 393s 1s/step - loss: 1.4391 - accuracy: 0.4717 - val_loss: 1.3638 - val_accuracy: 0.4990\n",
            "Epoch 3/30\n",
            "391/391 [==============================] - 401s 1s/step - loss: 1.2514 - accuracy: 0.5498 - val_loss: 1.1741 - val_accuracy: 0.5803\n",
            "Epoch 4/30\n",
            "391/391 [==============================] - 387s 990ms/step - loss: 1.1252 - accuracy: 0.5968 - val_loss: 0.9402 - val_accuracy: 0.6666\n",
            "Epoch 5/30\n",
            "391/391 [==============================] - 392s 1s/step - loss: 1.0258 - accuracy: 0.6346 - val_loss: 0.9384 - val_accuracy: 0.6695\n",
            "Epoch 6/30\n",
            "391/391 [==============================] - 387s 990ms/step - loss: 0.9677 - accuracy: 0.6589 - val_loss: 0.8876 - val_accuracy: 0.6883\n",
            "Epoch 7/30\n",
            "391/391 [==============================] - 388s 991ms/step - loss: 0.9079 - accuracy: 0.6814 - val_loss: 0.8476 - val_accuracy: 0.7070\n",
            "Epoch 8/30\n",
            "391/391 [==============================] - 396s 1s/step - loss: 0.8684 - accuracy: 0.6969 - val_loss: 0.8773 - val_accuracy: 0.6966\n",
            "Epoch 9/30\n",
            "391/391 [==============================] - 395s 1s/step - loss: 0.8477 - accuracy: 0.7040 - val_loss: 0.7635 - val_accuracy: 0.7341\n",
            "Epoch 10/30\n",
            "391/391 [==============================] - 399s 1s/step - loss: 0.8167 - accuracy: 0.7148 - val_loss: 0.7659 - val_accuracy: 0.7321\n",
            "Epoch 11/30\n",
            "391/391 [==============================] - 403s 1s/step - loss: 0.7909 - accuracy: 0.7264 - val_loss: 0.7056 - val_accuracy: 0.7552\n",
            "Epoch 12/30\n",
            "391/391 [==============================] - 407s 1s/step - loss: 0.7716 - accuracy: 0.7347 - val_loss: 0.7547 - val_accuracy: 0.7439\n",
            "Epoch 13/30\n",
            "391/391 [==============================] - 402s 1s/step - loss: 0.7595 - accuracy: 0.7370 - val_loss: 0.6631 - val_accuracy: 0.7747\n",
            "Epoch 14/30\n",
            "391/391 [==============================] - 398s 1s/step - loss: 0.7488 - accuracy: 0.7402 - val_loss: 0.6570 - val_accuracy: 0.7764\n",
            "Epoch 15/30\n",
            "391/391 [==============================] - 391s 1s/step - loss: 0.7268 - accuracy: 0.7505 - val_loss: 0.6517 - val_accuracy: 0.7764\n",
            "Epoch 16/30\n",
            "391/391 [==============================] - 402s 1s/step - loss: 0.7167 - accuracy: 0.7526 - val_loss: 0.6248 - val_accuracy: 0.7879\n",
            "Epoch 17/30\n",
            "391/391 [==============================] - 404s 1s/step - loss: 0.7088 - accuracy: 0.7552 - val_loss: 0.6784 - val_accuracy: 0.7688\n",
            "Epoch 18/30\n",
            "391/391 [==============================] - 395s 1s/step - loss: 0.6993 - accuracy: 0.7595 - val_loss: 0.6566 - val_accuracy: 0.7783\n",
            "Epoch 19/30\n",
            "391/391 [==============================] - 398s 1s/step - loss: 0.6907 - accuracy: 0.7642 - val_loss: 0.5855 - val_accuracy: 0.8003\n",
            "Epoch 20/30\n",
            "391/391 [==============================] - 394s 1s/step - loss: 0.6823 - accuracy: 0.7661 - val_loss: 0.6275 - val_accuracy: 0.7879\n",
            "Epoch 21/30\n",
            "391/391 [==============================] - 393s 1s/step - loss: 0.6731 - accuracy: 0.7686 - val_loss: 0.6382 - val_accuracy: 0.7817\n",
            "Epoch 22/30\n",
            "391/391 [==============================] - 401s 1s/step - loss: 0.6647 - accuracy: 0.7729 - val_loss: 0.6145 - val_accuracy: 0.7908\n",
            "Epoch 23/30\n",
            "391/391 [==============================] - 390s 997ms/step - loss: 0.6615 - accuracy: 0.7739 - val_loss: 0.6016 - val_accuracy: 0.7922\n",
            "Epoch 24/30\n",
            "391/391 [==============================] - 387s 990ms/step - loss: 0.6509 - accuracy: 0.7775 - val_loss: 0.5943 - val_accuracy: 0.7997\n",
            "Epoch 25/30\n",
            "391/391 [==============================] - 394s 1s/step - loss: 0.6396 - accuracy: 0.7820 - val_loss: 0.5965 - val_accuracy: 0.8006\n",
            "Epoch 26/30\n",
            "391/391 [==============================] - 396s 1s/step - loss: 0.6411 - accuracy: 0.7822 - val_loss: 0.6037 - val_accuracy: 0.7938\n",
            "Epoch 27/30\n",
            "391/391 [==============================] - 398s 1s/step - loss: 0.6301 - accuracy: 0.7850 - val_loss: 0.5551 - val_accuracy: 0.8132\n",
            "Epoch 28/30\n",
            "391/391 [==============================] - 397s 1s/step - loss: 0.6285 - accuracy: 0.7855 - val_loss: 0.6232 - val_accuracy: 0.7885\n",
            "Epoch 29/30\n",
            "391/391 [==============================] - 401s 1s/step - loss: 0.6226 - accuracy: 0.7888 - val_loss: 0.5651 - val_accuracy: 0.8111\n",
            "Epoch 30/30\n",
            "391/391 [==============================] - 392s 1s/step - loss: 0.6164 - accuracy: 0.7901 - val_loss: 0.5602 - val_accuracy: 0.8120\n",
            "10000/10000 [==============================] - 14s 1ms/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5ArymHUcx7Io",
        "colab_type": "code",
        "outputId": "141d98b4-70ff-4451-f3a9-92429b2bb91c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "import matplotlib.pyplot as pl1\n",
        "\n",
        "pl1.plot(history_adam.history['loss'],label='Adam',color=(0,0,1))\n",
        "pl1.plot(history_ams.history['loss'],label='AMSGRAD',color=(1,0,0))\n",
        "pl1.title('Training Loss')\n",
        "pl1.ylabel('Loss')\n",
        "pl1.xlabel('Epochs')\n",
        "pl1.legend()\n",
        "pl1.show()\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KAF_pmrfB7QB",
        "colab_type": "text"
      },
      "source": [
        "Có thể thấy AMSGrad hội tụ nhanh hơn ADAM trong phần lớn quá trình huấn luyện, tuy nhiên khi kết thúc huấn luyện cả hai thuật toán đều hội tụ về cùng một mất mát tương tự nhau. Thêm vào đó, tại một số lần lặp gần cuối, ADAM hội tụ nhanh hơn cả AMSGRAD. Những điều này không mang nhiều ý nghĩa vì đây chỉ là mất mát khi huấn luyện. Tuy nhiên, chúng lại khác với những kết luận được đưa ra trong bài báo 'On The Convergence Of ADAM And Beyond' khi huấn luyện với mô hình cifarnet, trong đó các tác giả kết luận răng train loss của AMSGRAD thấp hơn nhiều so với ADAM."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "T41MUoZyACVN",
        "colab_type": "code",
        "outputId": "91f2a852-0581-4a80-8104-4cf5cb766a78",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "import matplotlib.pyplot as pl2\n",
        "\n",
        "pl2.plot(history_adam.history['val_loss'],label='Adam',color=(0,0,1))\n",
        "pl2.plot(history_ams.history['val_loss'],label='AMSGRAD',color=(1,0,0))\n",
        "pl2.title('Validation Loss')\n",
        "pl2.ylabel('Loss')\n",
        "pl2.xlabel('Epochs')\n",
        "pl2.legend()\n",
        "pl2.show()\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "J8ifBYPQELor",
        "colab_type": "text"
      },
      "source": [
        "Validation loss lại cho ra kết quả khác xa Train loss. Có thể thấy, trong những lần lặp đầu, ADAM cho validation loss thấp hơn so với AMSGRAD. Ở những lần lặp sau đó, hai thuật toán liên tục đổi vị trí cho nhau khi so sánh giá trị validation loss."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Fij9lErfAK6b",
        "colab_type": "code",
        "outputId": "c14988dc-aa89-4b79-bf2f-14247cb759df",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "import matplotlib.pyplot as pl3\n",
        "\n",
        "pl3.plot(history_adam.history['accuracy'],label='Adam',color=(0,0,1))\n",
        "pl3.plot(history_ams.history['accuracy'],label='AMSGRAD',color=(1,0,0))\n",
        "pl3.title('Training Accuracy')\n",
        "pl3.ylabel('Accuracy')\n",
        "pl3.xlabel('Epochs')\n",
        "pl3.legend()\n",
        "pl3.show()\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7rme0V6wFMA8",
        "colab_type": "text"
      },
      "source": [
        "Training accuracy khá tương đồng với Training loss. AMSGRAD hội tụ nhanh hơn ADAM nhưng khi kết thúc quá trình huấn luyện, cả hai thuật toán đều đạt được độ chính xác tương đương nhau."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G2fQ3mt0AUBP",
        "colab_type": "code",
        "outputId": "b59bf419-746c-408e-f683-486b980a3521",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 295
        }
      },
      "source": [
        "import matplotlib.pyplot as pl4\n",
        "\n",
        "pl4.plot(history_adam.history['val_accuracy'],label='Adam',color=(0,0,1))\n",
        "pl4.plot(history_ams.history['val_accuracy'],label='AMSGRAD',color=(1,0,0))\n",
        "pl4.title('Validation Accuracy')\n",
        "pl4.ylabel('Accuracy')\n",
        "pl4.xlabel('Epochs')\n",
        "pl4.legend()\n",
        "pl4.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYkAAAEWCAYAAACT7WsrAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nO3dd3hUVfrA8e9L6AjSkQ4qkV4koggKoiIizcVFsAGKiIK6rrrWtcDPuq5iwYIKKgqsDQRFUBGRJhIUglRpShEMvSWGJO/vj3NDhjAzGSCTyUzez/PMk7n9vTNw37nnnHuOqCrGGGOMP0UiHYAxxpiCy5KEMcaYgCxJGGOMCciShDHGmIAsSRhjjAnIkoQxxpiALEmYqCUiKiJneu9fF5F/h7LuCRznWhH56kTjNCaaWZIwESMi00VkuJ/5PUVkm4gUDXVfqjpEVUfkQUz1vIRy5Niq+oGqdj7ZfQc5Zn0RyRSR18J1DGNOlCUJE0nvAteJiOSYfz3wgaqmRyCmSLgB2A1cLSIl8vPAIhKXn8cz0ceShImkyUAl4IKsGSJSAegGvCcibURkgYjsEZE/ROQVESnub0ci8o6I/J/P9L3eNltF5MYc614hIj+LyD4R2SQij/ks/t77u0dEDohIWxEZICJzfbY/X0QWiche7+/5Psu+E5ERIjJPRPaLyFciUjnQB+AlyBuAh4HDQPccy3uKyBIv1nUi0sWbX1FExnrnt1tEJnvzj4rVm+dbLPeOiLwmItNE5CBwUS6fByLSXkTme9/DJu8Y54jIdt8kIyJ/E5Glgc7VRCdLEiZiVDUF+BB3kczSB1ilqkuBDOAuoDLQFrgYuC23/XoX0nuAS4EGwCU5VjnoHbM8cAVwq4j08pZd6P0tr6qnqOqCHPuuCHwBvIRLcM8DX4hIJZ/VrgEGAlWB4l4sgbQHagETcZ9Ff59jtQHeA+71Yr0Q2OgtHgeUBpp4x3khyDFyugZ4AigLzCXI5yEidYEvgZeBKkBLYImqLgJ2Ar7FcNd78ZoYYknCRNq7wFUiUtKbvsGbh6ouVtUfVDVdVTcCbwAdQthnH2Csqv6iqgeBx3wXqup3qrpMVTNVNQmYEOJ+wV1Ef1XVcV5cE4BVHH0HMFZV1/gkwZZB9tcf+FJVdwPjgS4iUtVbdhMwRlW/9mLdoqqrRKQ6cDkwRFV3q+phVZ0dYvwAn6nqPG+fqbl8HtcA36jqBO84O1V1ibfsXeA6OJI8L/POwcQQSxImolR1LrAD6CUiZwBt8C40IhIvIp97ldj7gCdxdxW5qQFs8pn+zXehiJwrIrNEJFlE9gJDQtxv1r5/yzHvN6Cmz/Q2n/eHgFP87UhESgF/Bz4A8O5afsddmAFqA+v8bFob2OUllhPh+9nk9nkEigHgfaC7iJTBJeY5qvrHCcZkCihLEqYgeA93B3EdMENVt3vzX8P9Sm+gquWAB4Gcldz+/IG7uGWpk2P5eGAKUFtVTwVe99lvbt0ibwXq5phXB9gSQlw5XQmUA171EuE2XLLJKnLaBJzhZ7tNQEURKe9n2UFcMRQAInKan3VynmOwzyNQDKjqFmAB8DdcUdM4f+uZ6GZJwhQE7+HqDW7GK2rylAX2AQdEpCFwa4j7+xAYICKNRaQ08GiO5WVxv8RTvXL/a3yWJQOZwOkB9j0NiBeRa0SkqIhcDTQGPg8xNl/9gTFAM1yRVEugHdBCRJoBbwMDReRiESkiIjVFpKH3a/1LXHKpICLFRCSrLmUp0EREWnpFeI+FEEewz+MD4BIR6eOdbyUR8S0+ew/4l3cOn57AZ2AKOEsSJuK8+ob5QBncL9os9+AuWPuBN4H/hbi/L4GRwLfAWu+vr9uA4SKyH3gEl1Sytj2Eq9Sd57XmOS/HvnfiWl/djau4/RfQTVV3hBJbFhGpiauIH6mq23xei4HpQH9V/RFXAf4CsBeYTfZdzPW41lCrgD+Bf3jxrQGGA98Av+IqpnMT7PP4Hejqne8uYAnQwmfbSV5Mk7zPzsQYsUGHjDEnQ0TWAbeo6jeRjsXkPbuTMMacMBHpjavjyHm3ZmJEyN0eGGOMLxH5Dlcfc72qZkY4HBMmVtxkjDEmICtuMsYYE1DMFDdVrlxZ69WrF+kwjDEmqixevHiHqlYJtDxmkkS9evVITEyMdBjGGBNVRCRnDwJHseImY4wxAVmSMMYYE5AlCWOMMQHFTJ2EP4cPH2bz5s2kpqZGOpSYUrJkSWrVqkWxYsUiHYoxJsxiOkls3ryZsmXLUq9ePeSYETLNiVBVdu7cyebNm6lfv36kwzHGhFlMFzelpqZSqVIlSxB5SESoVKmS3Z0ZU0jEdJIALEGEgX2mxhQeMZ8kjDEmrA4dgjffhH37Ih1JWMR0nURBMnnyZK688kpWrlxJw4YNj1nesWNHnnvuORISEiIQnTHmhBw6BD16wMyZMHUqTJ4MRfLgt3dGBvvHT6Xs/q1w4EDur8aN4eOPT/64fliSyCcTJkygffv2TJgwgccffzzS4RhjTlZKCvTsCd9+C3//O3z0ETz1FDz00EntdscOSLzwXrqsfCF7pgiUKQOnnHL0q1IlqFvXJYlwUdWYeLVu3VpzWrFixTHzImH//v1ao0YNXb16tcbHx6uq6qFDh/Tqq6/Whg0baq9evbRNmza6aNEiVVUdMmSItm7dWhs3bqyPPPLIkf3UrVtX77//fm3RooW2bt1aFy9erJ07d9bTTz9dX3vttXw9p4Ly2RoTEYcOqV5yiaqI6nvvqWZmql5zjZueMeOEdpmZqTpunOo9ZV5VBf2szjCtwnbtcclB3bM7M49PIBuQqEGurYXmTuIf/4AlS/J2ny1bwsiRua/32Wef0aVLF+Lj46lUqRKLFy9m9uzZlC5dmpUrV5KUlMTZZ599ZP0nnniCihUrkpGRwcUXX0xSUhLNmzcHoE6dOixZsoS77rqLAQMGMG/ePFJTU2natClDhgzJ2xM0xhwrJSW7iOmdd+D669380aMhKQn69YPFi+E4OhzdsAGGDAH5ajqfczv7OnSjx8yRPPVOHEOGQNvzXWnWGWeE5YyCCmvFtYh0EZHVIrJWRO73s7yOiMwSkZ9FJElEuvose8DbbrWIXBbOOMNtwoQJ9O3bF4C+ffsyYcIEvv/+e6677joAmjdvfiQJAHz44YecffbZtGrViuXLl7NixYojy3r06AFAs2bNOPfccylbtixVqlShRIkS7NmzJx/PyphCKKuIaeZMGDsWbrghe1mZMvDpp5CeDlddBSE0E09Ph+eegyZNYM+cZUwp2Ye4ls0o9/kEiIvjppvg669h+3Y491z4/vswnlsAYbuTEJE4YBRwKbAZWCQiU1R1hc9qDwMfquprItIYmAbU8973BZoANYBvRCReVTNONJ5QfvGHw65du/j2229ZtmwZIkJGRgYiQqtWrfyuv2HDBp577jkWLVpEhQoVGDBgwFHPJJQoUQKAIkWKHHmfNZ2enh7ekzGmMEtJgV694JtvYMwY6N//2HUaNIBx41wiGTYM3nor4O5++gluvtn9vf7SbYxZ0Y2iWtbdMpxyypH1OnaEhQuhe3e45BJ4/XW48cYwnF8A4SxuagOsVdX1ACIyEegJ+CYJBcp5708FtnrvewITVfUvYIOIrPX2tyCM8YbFxx9/zPXXX88bb7xxZF6HDh1o3bo148ePp1OnTvzyyy8kJSUBsG/fPsqUKcOpp57K9u3b+fLLL+nYsWOEojcxKS3N/Qru2hVq1450NMdQdSEeOACZmVAl4EgHPvbtg19/Dd4C6OBB9zctDS6+GP72Nzj11IC7nD0bNm6Ebt2gUplUuPJK97P+7bdhwIDAsfTo4Sqvn3gCzjsPBg06avGhQ/Doo/DCC1C5Mnwy7hBXvtQD2b0D5s6FWrWO2eWZZ8KCBdCnD9x0E6xcCU8/DXFxIXw2JymcSaImsMlnejNwbo51HgO+EpHbgTLAJT7b/pBj25o5DyAig4HB4MrqC6IJEyZw3333HTWvd+/e/Pzzz6SkpNCoUSMaNWpE69atAWjRogWtWrWiYcOG1K5dm3bt2kUibBOrVN1V5v33oWxZd6UZMiRvmm3mkJrqikm2b4dt27Jf27fDrl3Br+e+N8VDh7oL6jFdhanCnDnuov3RR+6XfiBFi2a3CEpPhw8+gFtvdRngmmtcwixZEnC55J573C92gDJxqXxXoRetd35FyktvUXrgwNxP/vHH4ccfXfAtWsA55/D7764h1PDhrg5i0CB49ulMKgy+HhITXfPZACUMAOXLw7RpcOedrohq9Wp3GmXL5h7OSQlWq30yL+Aq4C2f6euBV3Ks80/gbu99W9xdRhHgFeA6n/XeBq4KdryC3LopFtlnG6UeekgVVP/5T9VLL3Xv27VTPcnvc+9e1VtuUe3YUbVhQ9Xy5d2u/b0qVVJt0ED17LNVL7xQtWtX1T59VG+8UfWOO1QffFD1ySdVX3pJ9dZb3TYdOqj++ad3sD/+UH36adX4eLewbFnVwYNVP/1UdeZM1YULVZcvV/3tN9UdO1RTU13ToSyZmao//OAOVq2a20e5cqoDB+ryF7/W+DPSVUT1nntUf/w+RVfV76IKOpC3tUQJ1V69VCdMUD1wIPDnkZmpuvaHZN1fsY7uKF1bW9ZKPnL+8fGq333nrXjvvW7mCy8c1+f9yiuqcXGqzZurbtx4XJseg1xaN4UzSbQFZvhMPwA8kGOd5UBtn+n1QNWc6wIzgLbBjmdJIn/ZZxuF3njD/ZcfNMhdxTIzVd95R7VCBdXixVUff1z1r79OaNfXXecuWhdcoHrVVarDhqn+3/+pvvWW6uefqy5apLpp04ntftw41dLFD+uNVabong493IFAtX171bFjg1+tc3P4sOpXX2nGDQM0pXhZVdDtcafppt53qs6bp3r55aqgmW++pQsWqN55p2qNGu7wpUu75Pbpp6oHD6ouW6Y6apSbd9ppbp3WLNIUSujSqpfoyyPTdelS1YwM79ijR7uVbrvt6CQWohkzVE89VbVqVdX580/8I4hkkijqXfTrA8WBpUCTHOt8CQzw3jfC1UkIrsJ6KVDC2349EBfseJYk8pd9tlHmiy/cxbVLF3dh9LVtm2rfvu5y0KSJ6oIFx7XrDz5wmz7+eB7Gq6qakuLucB54QP+qXN1dwKmqK3vcq7pyZZ4dZsUKd1dTkkP6csePNa3blS5pZv30f/PNo9bPyFCdPdvd5VSp4lYRyV69Vi3Va691OXnlStXMN99yCx54IHsnX33lvo/LLz/2+zjO2M84Q7VpU9X09BPbR8SShDs2XYE1wDrgIW/ecKCH974xMM9LCEuAzj7bPuRttxq4PLdjWZLIX/bZRpHERNUyZVRbtVLdty/welOnuiuciCuK2b8/111v2OBKatq1O45r3cqVqmPGqD73nLtwDh7sbj8uuki1RQsXQ6lS2VfdIkVUu3XTXWMm6QXnpSm4UrMjv8hPUEaG6osvqpYsqVq5srsjOGL3bhfj5MlB9+HdiOj997ubsvXrA9wUDBrkzmXSJNVffnEfWrNmrpzuJO3Yofrrrye+fUSTRH6+LEnkL/tso8T69a7cvW5dTd2wVSdNUv3b31RPOcWVdhxj717VoUNdoqhTR3XatIC7PnzYJYdy5VyyyNUPP7gCfd8KiqJFXXlJo0ZuZz16qA4YoHr33a5i4s03VTdvPrKL1FTVm25ym3bvfuLX2E2bVC++2O2nWzdXzRFWKSmqCQnuw6pTx5VH/fZbmA8aGksSJizssz0J+/a52tk77lBNSwvfcXbu1MyzztLDZcvrY1ev0AoV3P/4qlXdTYWIKyrya+5cd+HOuop+8IHqnj1HrfL4425xwH2oup/VM2a4uwRw9R+PPOJ++u7de0Jl8ZmZ2RW3jRqprlkT+rY7d7pf/OXLu5ur0aNPKIQTs3Gjq7UvVcpV0hQQliRMWNhnewIyM1XHj8+u+QTV3r1zTRS//ab6++/HV+m79pcU3Vi7vaZSXC9gtpYqpdqvn7sxOHzYdT100UXuQhuwRCU11WWC6q4+QIsVc2Xob76pP37xp8bFubJ3v9LTVT/6SLV1a7dtjRqq//1vSEVYoZo1y11zy5dXnT49e/7hwy4HTZ3qSrQGDXJ13Fn1B6B6/vmqa9fmWSihW71adenSCBw4MEsSBcSkSZMU0JVehduGDRsU0IceeujIOsnJyVq0aFEdOnSoqqquWrVKO3TooC1atNCGDRvqzTfffGTdhQsXaocOHfTMM8/UVq1aadeuXTUpKUlVVR999FGtUaOGtmjRQhs1aqTjx48PGktWPCVLltSWLVtqw4YN9ZxzztGxY8cGPJ+C9NlGhaQk144TXC3pggWqzz/vpq+80m8GyMjIbrGa9apYUbVxY9VOndxF/667VJ95RvXdd90P9lGjVNuem6ET6aMKOqLJBH3nHf9VEfv2qZ57rquj/frrILFnZLjmM/fco1q/vipoOkV0fsmOeuiZl1zZTZbUVNekKat5any8m05NPemP0J8NG1wz0CJFXP5q0uToOuesO6cLLlC9+WaXp7788qTqimOOJYkCok+fPtq+ffsjvbpu2LBB69evry1btjyyzquvvqotWrQ4kiQ6d+6sk31+5mUlgW3btmndunV13rx5R5bNmTNHJ02apKouSfznP/9RVdU1a9Zo2bJlNc3n12rOWLLiadKkyZHpdevWaYsWLXTMmDF+z6cgfbYF2u7drt1kXJy7wr/xxtHNUF580f037NHjqAvp/v3ZxfcDBrhikREjXHVB796u+P6MM1yRSc7nEMZUvkcVdPdDz+Ya3q5d7iJburQrYcpVZqY+3O1nHc6/9UD9JtkHbdPGZayaNbMT4UcfnXiTm+Nw4IBq//7u+YyePVXvu8/VOc+f74qXTHCWJAoAf12FZ12U+/Xrd6SL8A4dOugTTzxxJEk0a9ZMExMTj9nfww8/fNQFPiffJKGqWq1aNd2+fXvAWHzj8TVz5syjkpivgvLZFiQHDrhfqqtWqfv1PXas+xkrojpkiGuG4s8rr7j/ildcoZqaqhs3Zv86Hjky9zLz/ftd0cncuapbHnjZ7Wvo0JAL27dtcz/4y5VTXbw4+LoTJrjdP/qoN2PVKtWnnlI95xy3oFMn19wn3wr6zcnKLUkUmq7CI9lXuL+uwitVqgS4XmEnTpxItWrViIuLo0aNGmzd6rqwuuuuu+jUqRPnn38+nTt3ZuDAgZQvX57ly5fT31/nYn789NNPNGjQgKpVqwaMJatLkJzOPvtsVq1aFdJxCruNG12fbklJMPFfP/G/qsOo/8cC13fPl1+CT1fwxxg61HXCc+ut7Op4Je3Xfsr+wyWZNg0uC6H/41NOgVNKZXDGzLfh6TtcIC++6AaqCUG1aq7PugsucMebPdv/GDa//eZ68GjbFh5+2Jt51llw//3ulZICpUqFdEwTPQpPkoigCRMmcOeddwLZXYUPGzYMgC5duvDvf/+batWqcfXVVx+13cCBA7nsssuYPn06n332GW+88QZLly49Zv/nnnsu+/bto3Pnzrz44osAvPDCC4wdO5Y1a9YwderUoLEEShLuR0Yhk5LiOhf688+j/x44EHCT336DSZPghkzo3X4TdeZ+QPIfVRhWZixNrruBm5sXyf0/2pAhzFtQhHbv3cIHpXtx2vxJxLcI4YKbng7jx7vO5NascVf68eOPu+e32rWzE8Wll7oukU4/PXt5RoYbNiEz0/UXVNTfCVmCiE3BbjOi6VVQi5t27typpUqV0jp16mjdunW1Vq1aWrt2bV2/fv2R4p2BAwdqtWrVdOfOnTp27NgjxU05NWnSRBMTE/0WN3300Ufav39/VT26uOmzzz7TGjVqaEpKSsBYMjMzAxY3tWrVym8sBeGzPSmpqa5WuFcv1bZtXQH/KafoMQX8vu35ixU76pVZrJimxxXTv3CvzGLFXOH+nXdq0ve7j9RTN22q+s03gUNJT3ddKYHqc43e0kwR16/SwYOBN/rrL1chfPrpbsOWLVU/+eSknzBbtsxVndSvf9TjCTpihDvMuHEntXtTAJFLcVNYBx0y2V2F//bbb2zcuJFNmzZRv359Nm3K7iD37rvv5plnnqFixYpHbTt9+nQOHz4MwLZt29i5cyc1a9Zk6NChvPPOO8yfP//IuocOHfJ7/B49epCQkMC7774bMJY5c+Ycs93GjRu55557uP322/PiYyh4nn46+9d3qVLQpo3rHfXJJ90YAFOnul48N250fTsfPuy6mPZef+1P4+Yb0iiakUbvbmmk7k1D0tJcF6IjR9LsgvLMmuXGpj9wwI0D0KsXrF17dBh797pxAp5/Hm6/He5MugkZM8b9rO/e3R3b119/wRtvQHy860a0YkWYMsUNSvC3v510b65Nm8KMGW6c5UsugeRkN5bBY4+5Adeuvfakdm+iUbAMEk2vgnon0bFjR/3yyy+Pmvfiiy9qly5djvnlrqpH3UncddddGh8fr82bN9fmzZvrOJ+fcQsWLNALL7xQzzjjDG3btq127979SAV4zorrxMREjY+PDxjLkCFDClcT2FWrXDvJvn1PaPM//nA3H6F2D5GS4h4eLlPG3YTce697juzXX93DYEWLqr7+eo6N3n3XVXh37OhqxFNSVF9+2XVZAarnneceeghTBfH337tnvlq2dDdZdeu6hlom9mCtm0w4RO1nm5npLrzly59QXww//uhaeZYurfrhh8e37datrjlrVtv9ChVc0c6sWQE2eP9918SpVavsB9rat3cPNeRD66Hp011SK1JEdc6csB/OREhuScIqrk3h8s478N13rsjmtNOOa9P333clPKedBvPnu7Fkjkf16m5AuKFD4Z//hP374ZNPjq4gPsq117rioxtuyK6Q7tAh5FZLJ+uyy9xAbDt3Qvv2+XJIUwCJSyTRLyEhQRMTE4+at3LlSho1ahShiGJbQfxsMzLcsMKLFsHdd7uhHo9q5JOcDA0bQqNGbkT5EMvvMzJcC8/nnnPjDX/4YYhDauaVQ4egdOl8PKApTERksaomBFoe8xXXsZIEC5KC+Jmmp7sf3K+/7lqsXnONywdvv+3qmgGXOfbvh9GjQ0oQu3a5xw2aN3cJYtgw+OqrfE4QYAnCRFRMJ4mSJUuyc+fOAnlRi1aqys6dOynpjQdcEBw+7FrejB8PTz3lnlv4+GM39u+gQW4Q+cnDvoFx4+Bf//L/pJhHFWbNckmmRg33DGaZMjBxIrz8sp9xlo2JcTFdJ1GrVi02b95McnJypEOJKSVLlqRWrVqRDgNwLUL79HGtQJ9/Hu66y83v3du1CJ0+Hf4zPIUmo25lfdyZTC71EDfvP3bw+G3bXHXF22+7Zqrly8PNN7skc7x1D8bElGC12tH08te6ycS2Q4eODEGso0YFXi/zQdeV6r1nf3NkSIPHHlNNTnajel55pWuGCqoXXugeGDt0KP/Ow5hIIpfWTTFdcW1i18GDrouib791VQyDBgVYcfly18dWv37w3nv8+KN7hm7KlOxVqlSB/v3dPs46K1/CN6bAyK3iOqzFTSLSBXgRiAPeUtWncyx/AbjImywNVFXV8t6yDGCZt+x3Ve0RzlhNAbZ4MVxxBXTtCk89xf7S1ejWDebOdUVEN9wQYLvMTLjlFjj1VPjvfwH3YPVnn8GyZa5J6znnQI8eULx4vp2NMVElbHcSIhIHrAEuBTYDi4B+qroiwPq3A61U9UZv+oCqnhLq8exOIkZt2eKu7GlpsHcvWqoUL1d4lPs2DWPsB8Xp2zfItqNHuyQxdiwMGJBfERsTVSLZBLYNsFZV16tqGjAR6Blk/X7AhDDGY07C77+7Rwvy1aFDrkxp3z6YOZO9835hnrTnjt/uZnv1FvStMCPwttu2uZZMHTu6siRjzAkJZ5KoCWzymd7szTuGiNQF6gPf+swuKSKJIvKDiPQKX5gmFIMHu+vtpEn5dMDMTHdx/+knGD+eHTWa03FwPBenfMHCf39OudIZ0KWLSyLr1h27/V13uW6/X389355QNiYWFZTnJPoCH6tqhs+8ut4t0DXASBE5I+dGIjLYSySJ1sw1fPbsgZkz3RgC117regUNu8cfh48/Rp/9D1+X7E6HDrBqlatPOHf4Fa5S4ZlnXM1148bw4IPZYz5Mn+4ebHjoIauJNuZkBWv6dDIvoC0ww2f6AeCBAOv+DJwfZF/vAFcFO541gQ2fDz5wzUMnTXLDF1SporpuXRgPOH68KuiGi2/UhNaZCqo1aqjOnOln3S1bVK+/Xo+sNHasar16bsBjnzGjjTH+EcHxJBYBDUSkvogUx90tTMm5kog0BCoAC3zmVRCREt77ykA7wG+Ftwm/yZPdEJc9esC0aa4LjK5dXbcVeS193kIy+g9kUekLiZ/5Grv3CG++CevXQ6dOfjaoUQPee8/1uFe9Ogwc6MaAeOMNKFEi7wM0ppAJW5JQ1XRgGDADWAl8qKrLRWS4iPg2Z+0LTPQyWpZGQKKILAVmAU9rgFZRJrxSU90QzT17uu6OzjrLJY0NG9wTzX/9lTfHSUmB957YxO4Le/Lb4RrcXfcT3ptQnFWr3PMLuV7v27Z1gwS9845LEBdemDeBGVPI2cN0JqjPP3cDpE2bBpdfnj1//HhXP3Htta5LpBOtG963z9Utv/7cAT5Nbs+ZcRtIfGkBHW5tbPXNxuSDiD5MZ6Lf5Mmun6OcRT3XXOPuJh5+2I2HMHz48e13714YOdK99u7JZE7V62lRZBlM/YKOlwfugM8Yk78sSZiAMjJc9xVXXOG/uOfBB11dwYgRLlGE8rzagQOuN9X//Ad273bjPr9a/mGqvzPZ9ct9eZc8Pw9jzImzJGECmj/fjdPTK8BTKiKuqOj3312PqbVrw8UdM9zgC6pQs6arWK5Zk0Pla/DaB+V45hm3z27d3N1Hq1/GwQ1PuSejb789f0/QGJMrSxImoEmTXJ9GvnURORUr5sZuaN/eVWQv/9f71Hr4/mPWKw0M5hT6lK5JudY1OLV8TXivMrz6qivLevlle+jNmALIkoTxS9XVR1x8MZQrF3zdU0+FL76Adm0Ok/no4xxu2gpmz+bTUX/w8UtbKb5jC+1P30qPswb/dIsAACAASURBVLdQW7bC1q2ud76tW92DcB99ZKP5GFNAWZIwfiUluYrpBx4Ibf06dWDuoLHUeWIDt+77ghkJZdmwoSxt28YzYqK7WTjmRiGrZZ3dQRhTYBWUbjlMATNpkrt29wi1g/bUVOq+O4JdDdvy5ubLqVjRNZudN8/djfjNAyKWIIwp4OxOwvg1eTKcf7570joko0fD5s1UnPkuf7YUKlSw678xscDuJMwxNmyApUvhyitD3ODgQXjySbjoIujUiYoVLUEYEyvsTsIcY/Jk9zdQ09djjBoF27fDJ5+ELSZjTGTYnYQ5xqRJ0KwZnHFM5+x+7Nvnuuy+/HJo1y7ssRlj8pclCXOUP/90lc0h30WMHOm6gx0xIqxxGWMiw5KEOcrUqW5QuJDqI3btgv/+163cunXYYzPG5D9LEuYokyZB3brQsmUIKz/3HOzf70aRM8bEJEsS5oj9++Gbb1xRU66tk/7803XI17evq8AwxsQkSxLmiOnT3SBCIRU1Pf20G5HoscfCHZYxJoIsSZgjJk+GSpVCaKS0ZYvrmK9/f4iPz5fYjDGRYUmiMFu3zvWZ8eSTpO0+yBdfuG44iub29MwTT7ja7UceyZcwjTGRY0misFq1yo0DPX8+PPQQmWc2oM/e0fytR3rw7TZsgLfecgNP16uXL6EaYyInrElCRLqIyGoRWSsixwwyICIviMgS77VGRPb4LOsvIr96r/7hjLPQ+eUX6NAB0tNh0SKYM4ctxeozmlvoel9T+PTT7B5acxoxAooUgYceyt+YjTEREbYkISJxwCjgcqAx0E9Ejhq8WFXvUtWWqtoSeBn41Nu2IvAocC7QBnhURCqEK9ZYkZkJX34JPXvCpZfCkiV+VvrpJ+jY0ZUpzZ4NTZuSeX57LpC5PN32M4oUjYPevaFtW7fc15o18O67cNttbtQ5Y0zMC+edRBtgraquV9U0YCLQM8j6/YAJ3vvLgK9VdZeq7ga+Bmzw4wB27HBjRjdoAF27wsKFbjyIhAT3gz811Vtx4UI3sEOZMvD999CwIQA//gh/bBNq3dbD9ez39tuucrpjRzfAdVKS2/6xx6BkSbj/2JHnjDGxKZxJoiawyWd6szfvGCJSF6gPfHs824rIYBFJFJHE5OTkPAk6Wqi6a37//lCrFvzrX26M6f/9z405vXIl3HCD65y1ZUtIGjUHLrkEKld2CcKnY6ZJk9yNxRVX4N7ceKO7a3j2WVdn0bIlXHUVTJwId94JVatG7sSNMfmqoFRc9wU+VtWM49lIVUeraoKqJlSpUiVMoRUshw65H/oJCXDeee4CP2iQq2b47jvo08eNS12xIowZA199Ba33zOSMYV3YVrQm+z+f7R6p9qi6fVx0EVTwLdArVQruvRfWr3cZ6Isv3Dim99yT7+dsjImccCaJLUBtn+la3jx/+pJd1HS82xYK6elw332uKmDQIEhLg9dec6VCr7wCTZr43+7S9C95f88V7K10Oi33zKZJ55pMm5a9fOVK+PXXIB36VajgHpxbvx4SE132McYUGuFMEouABiJSX0SK4xLBlJwriUhDoAKwwGf2DKCziFTwKqw7e/MKrQcfdKU/nTvDnDmummDIEChbNshGn30GPXsijRtTY9UsJi+oRtmyrljpuutcXUbW2BE9g9UWAVSvDmeemVenY4yJEmEbdEhV00VkGO7iHgeMUdXlIjIcSFTVrITRF5iomt3mUlV3icgIXKIBGK6qu8IVa0E3ebKrmB4yxN09hOTDD+Haa13vrF9+CRUqcF5l17jpqadcXcWMGVC6NLRpY42VjDH+iQZqDx9lEhISNDExMdJh5Lm1a911Pj4e5s6FEiVC2GjcOBgwwA1SnVWXkMMvv8BNN7mWTU8/7YqyjDGFj4gsVtWEQMtt+NICLCXFNSqKi4OPPgoxQbz9Ntx8s6uJnjLFNXf1o2lT13Dpm29cS1djjPHHkkQBNnSoq3v44osQe8B49VW30WWXuSZLpUoFXT0uzq1qjDGBFJQmsCaHt9+GsWPh4Yfd8NG5GjnSJYju3V0lRi4JwhhjQmFJogD6+Wd3vb/0Unj00RA2eOYZuOsu153Gxx+7p6KNMSYPWJIoYHbvdtf6KlXggw9ckVBQI0a4bjL69nVPRBcvni9xGmMKB6uTKEAyM103G5s2uZ4zgj5Ergr//rcb2+GGG9zj1blmFGOMOT6WJAqQZ5+FqVPd0NFt2wZZUdV1lfHcc+7x6zfecN13G2NMHrMrSwExa5brsbVPH7j99iArqsI//uESxG23WYIwxoSVXV0KgK1bXZVCfLwb9E0kwIqZmXDrrfDSS66i+pVXLEEYY8LKipsi7PBhuPpqOHAAvv3WT19MaWluLOo1a1zF9MSJrqL6ySeDZBNjjMkbliQi7MknYe5c5dOXttBk+xr4frVLCGvWwOrVbkzpzEy3sgg8/rirsLYEYYzJB7kmCRHpDnyhqpn5EE+hsncv7H76DQ7F3U2pOw5mLyhd2pU9tW4N/frBWWe56fh4KF8+cgEbYwqdUO4krgZGisgnuJ5cV4U5pkLjtdegd+r7FKlVDR68JzsZ1KhhdQ3GmAIh1yShqteJSDncGNTviIgCY4EJqro/3AHGqpQUeOX5NP5ZZBHF/36bq5A2xpgCJqSfq6q6D/gYmAhUB64EfhKRYI01TRBjx0KN5CUUz/zLdeltjDEFUK5JQkR6iMgk4DugGNBGVS8HWgB3hze82JSe7gYRuqaeNxhf0CfnjDEmckKpk+gNvKCq3/vOVNVDInJTeMKKbRMnwsaN8PcLFkBGbRsWzhhTYIWSJB4D/siaEJFSQDVV3aiqM8MVWKzKzHQjwTVtCjU2zreiJmNMgRZKncRHgG/z1wxvXq5EpIuIrBaRtSJyf4B1+ojIChFZLiLjfeZniMgS7zXF37bR6PPPYflyeHzwFmTTJitqMsYUaKHcSRRV1bSsCVVNE5Fc+6MWkThgFHApsBlYJCJTVHWFzzoNgAeAdqq6W0Sq+uwiRVVbhnoi0UAVnnoK6teHnlWtPsIYU/CFcieRLCI9siZEpCewI4Tt2gBrVXW9l2QmAj1zrHMzMEpVdwOo6p+hhR2dZs+GH36Ae++FuIXz3eBALWMqDxpjYkwoSWII8KCI/C4im4D7gFtC2K4msMlnerM3z1c8EC8i80TkBxHp4rOspIgkevN7+TuAiAz21klMTk4OIaTIeuopqFYNBg4EFiyAhAQbJMgYU6CF8jDdOuA8ETnFmz6Qx8dvAHQEagHfi0gzVd0D1FXVLSJyOvCtiCzzYvGNbTQwGiAhIUHzMK48t3gxfPWVq7QuSSr89BPceWekwzLGmKBC6uBPRK4AmuB+3QOgqsNz2WwLUNtnupY3z9dmYKGqHgY2iMgaXNJYpKpbvOOsF5HvgFbAOqLU00/Dqad6D1b/9JPr3dVaNhljCrhQHqZ7Hdd/0+2AAH8H6oaw70VAAxGp71V09wVytlKajLuLQEQq44qf1otIBREp4TO/HbCCKLV6NXzyCQwdCuXK4YqawCqtjTEFXih1Euer6g3AblV9HGiLu5gHparpwDBgBrAS+FBVl4vIcJ+K8BnAThFZAcwC7lXVnUAjIFFElnrzn/ZtFRVtnn0WSpTwKV1asMA1capWLaJxGWNMbkIpbkr1/h4SkRrATlz/TblS1WnAtBzzHvF5r8A/vZfvOvOBZqEco6DbvBnGjYNbboGqVXHtYOfPh06dIh2aMcbkKpQkMVVEygP/AX4CFHgzrFHFkP/+1+WFe+7xZvz+O/zxhxU1GWOiQtAkISJFgJlea6NPRORzoKSq7s2X6KLcjh0wejRccw3UzarFsfoIY0wUCVon4Y1GN8pn+i9LEKF7+WU4dAjuu89n5vz5buS55s0jFpcxxoQqlIrrmSLSW8QGVT4e+/e7JNGrFzRu7LNgwQJo0waK2vDixpiCL5QkcQuuQ7+/RGSfiOwXkX1hjivqjR4Nu3fDAw/4zDx0CJYssaImY0zUCOWJ67L5EUgsSUuD5593DZjatPFZkJjoRhyyJGGMiRK5JgkRudDf/JyDEJlss2fD1q3w6qs5FliltTEmyoRSMH6vz/uSuN5dFwPW0D+AqVOhVCm49NIcCxYsgAYNoHLliMRljDHHK5Tipu6+0yJSGxgZtoiinCpMmQKXXOIaMR21YMEC6NIl4LbGGFPQhFJxndNmXLcZxo9ffoHffoMePXIsWL8e/vzTOvUzxkSVUOokXsY9ZQ0uqbTEPXlt/Jg61f294oocC6w+whgThUKpk0j0eZ8OTFDVeWGKJ+pNmQLnnAPVc/ZutWABlC0LTZpEJC5jjDkRoSSJj4FUVc0AN3a1iJRW1UPhDS36bNsGP/4Iw/2NtDF/Ppx7LsTF5XtcxhhzokJ64hoo5TNdCvgmPOFEty++cPXT3bvnWHDgACQlWVGTMSbqhJIkSvoOWeq9Lx1k/UJr6lSoU8dPt0yLFkFmpiUJY0zUCSVJHBSRs7MmRKQ1kBK+kKJTSoobw7p7dziml6v5893f887L97iMMeZkhFIn8Q/gIxHZihu+9DTccKbGx7ffukRxTFETuErrRo2gQoV8j8sYY05GKA/TLRKRhsBZ3qzVqno4vGFFn6lT4ZRToGPHHAuyHqLr1SsSYRljzEnJtbhJRIYCZVT1F1X9BThFRG4LZeci0kVEVovIWhG5P8A6fURkhYgsF5HxPvP7i8iv3qt/qCcUCaouSVx2mRvL+ihr1sCuXfYQnTEmKoVSJ3GzNzIdAKq6G7g5t41EJA43YNHlQGOgn4g0zrFOA+ABoJ2qNsEVbSEiFYFHgXNxfUU9KiIFtqzmp59ch37HPGUN9hCdMSaqhZIk4nwHHPIu/sVD2K4NsFZV16tqGjAR6JljnZuBUV7iQVX/9OZfBnytqru8ZV8DBbbToylToEgR6NrVz8IFC6B8eWjYMN/jMsaYkxVKkpgO/E9ELhaRi4EJwJchbFcT2OQzvdmb5yseiBeReSLyg4h0OY5tEZHBIpIoIonJyckhhBQeU6e60iS/nbvOn+9aNRU5kW6yjDEmskK5ct0HfAsM8V7LOPrhupNRFGgAdAT6AW+KSPlQN1bV0aqaoKoJVapUyaOQjs+mTfDzzwFaNe3dC8uXW1GTMSZq5ZokVDUTWAhsxBUhdQJWhrDvLUBtn+la3jxfm4EpqnpYVTcAa3BJI5RtC4TPP3d//SaJH390tdqWJIwxUSpgkhCReBF5VERWAS8DvwOo6kWq+koI+14ENBCR+iJSHOgLTMmxzmTcXQQiUhlX/LQemAF0FpEKXoV1Z29egTN1Kpx5ZoAqhwUL3JN1556b73EZY0xeCPacxCpgDtBNVdcCiMhdoe5YVdNFZBju4h4HjFHV5SIyHEhU1SlkJ4MVQAZwr6ru9I41ApdoAIar6q7jPLewO3AAZs6EoUP9PGUNrj6iaVMoVy7fYzPGmLwQLEn8Dffrf5aITMe1TvJ3KQxIVacB03LMe8TnvQL/9F45tx0DjDme4+W3r7+GtLQATV8zM+GHH+BqezjdGBO9AhY3qepkVe0LNARm4Z5hqCoir4lI5/wKsCCbMsW1bm3Xzs/CVatcxbXVRxhjolgoFdcHVXW8N9Z1LeBnXIunQi0jw3UNfvnlUKyYnxWyOvWzJ62NMVHsuBrvq+pur9npxeEKKFosXAjJyQGKmsBVWleqBA0a5GtcxhiTl+wJrxM0dSoULQpdAj0HvmCBe4jOb422McZEB0sSJ2jqVLjwQlcncYxdu2DlSquPMMZEPUsSJ2D9evcgtd8H6CC7PqJ9+3yLyRhjwsGSxAmYOtX9DZgk5sxxtdlt2uRbTMYYEw6WJE7AlCnQuDGccUaAFebMgXPOgVJ51cWVMcZEhiWJ47RnD3z/fZC7iJQUSEyECy7I17iMMSYcLEkcp+nTIT09SNPXhQvh8GFLEsaYmGBJ4jhNnerGjQjYZ9+cOa7Zqz1EZ4yJAZYkjsPhwzBtGnTrBnFxAVaaM8d16lehwI62aowxIbMkcRzmzXN1EgHrI9LT3UN0VtRkjIkRliSOw7RpULw4XHppgBWWLnX9h1uSMMbECEsSx2HxYmjRAsqWDbDCnDnuryUJY0yMsCQRIlVISoLmzYOsNGcO1K8PNWvmW1zGGBNOliRCtH077NgRJEmouiRhdxHGmBhiSSJESUnub8AksWaN6zvc+msyxsSQsCYJEekiIqtFZK2I3O9n+QARSRaRJd5rkM+yDJ/5U8IZZyiykkSzZgFWsPoIY0wMCjbG9UkRkThgFHApsBlYJCJTVHVFjlX/p6rD/OwiRVVbhiu+45WU5KoaKlUKsMKcOVClCpx1Vr7GZYwx4RTOO4k2wFpVXa+qacBEoGcYjxdWy5YFuYsAlyTat7dBhowxMSWcSaImsMlnerM3L6feIpIkIh+LSG2f+SVFJFFEfhCRXv4OICKDvXUSk5OT8zD0ox0+DCtWBKmP2LIFNmywoiZjTMyJdMX1VKCeqjYHvgbe9VlWV1UTgGuAkSJyTMfc3njbCaqaUKVKlbAFuWYNpKUFSRJz57q/liSMMTEmnEliC+B7Z1DLm3eEqu5U1b+8ybeA1j7Ltnh/1wPfAa3CGGtQubZsmjMHypSBlgWmCsUYY/JEOJPEIqCBiNQXkeJAX+CoVkoiUt1nsgew0ptfQURKeO8rA+2AnBXe+SYpyQ00F7BOes4cN5510bC1AzDGmIgIW5JQ1XRgGDADd/H/UFWXi8hwEckajeEOEVkuIkuBO4AB3vxGQKI3fxbwtJ9WUfkmKQkaNnT9Nh1jzx5Xq21FTcaYGBTWn76qOg2YlmPeIz7vHwAe8LPdfCBYW6J8tWwZXHhhgIXz5rmnrS1JGGNiUKQrrgu83bth06Zc6iOKFQsyCpExxkQvSxK5WLbM/Q3asql1ayhdOt9iMsaY/GJJIhdBWzalpsKiRVbUZIyJWZYkcpGUBBUrQvXqfhb++KN7gMI69TPGxChLErlYtszdRfjtbSOrU7927fI1JmOMyS+WJILIzMxOEn7NmQNNmgTp9c8YY6KbJYkgNmyAgwcDJImMDJg/3+ojjDExzZJEEEErrZcuhf37LUkYY2KaJYkgkpJcXUSTJn4W2iBDxphCwJJEEElJcOaZAR6BmDsX6taF2rX9LDTGmNhgSSKIgJXWqtmDDBljTAyzJBHAwYOwdm2AJLF2LWzfbkVNxpiYZ0kigOXL3Q2D3yRh9RHGmELCkkQAQVs2zZnjno1o1ChfYzLGmPxmSSKApCQ32Fy9en4WZtVH+H0M2xhjYocliQCWLYNmzaBIzk/ojz9g3TorajLGFAqWJPxQdXcSVh9hjCnsLEn4sXUr7NoVIEnMnesenGjVKt/jMsaY/BbWJCEiXURktYisFZH7/SwfICLJIrLEew3yWdZfRH71Xv3DGWdOuVZan3eeG43OGGNiXNiShIjEAaOAy4HGQD8Raexn1f+pakvv9Za3bUXgUeBcoA3wqIhUCFesOWUliWY5R9neu9f12WRFTcaYQiKcdxJtgLWqul5V04CJQM8Qt70M+FpVd6nqbuBroEuY4jxGUpLrbaN8+RwL5s93FRaWJIwxhUQ4k0RNYJPP9GZvXk69RSRJRD4WkayOkELaVkQGi0iiiCQmJyfnVdyBu+OYPRuKFnXFTcYYUwhEuuJ6KlBPVZvj7hbePZ6NVXW0qiaoakKVKlXyJKC0NFi50k+SyMiACRPgoovcAxTGGFMIhDNJbAF8u0it5c07QlV3qupf3uRbQOtQtw2XVasgPd1PkvjmG/j9dxg0yO92xhgTi8KZJBYBDUSkvogUB/oCU3xXEJHqPpM9gJXe+xlAZxGp4FVYd/bmhV3Alk1vv+264ugZarWKMcZEv6Lh2rGqpovIMNzFPQ4Yo6rLRWQ4kKiqU4A7RKQHkA7sAgZ42+4SkRG4RAMwXFV3hStWX0lJULw4NGjgM3PHDpg8GYYOhRIl8iMMY4wpEMKWJABUdRowLce8R3zePwA8EGDbMcCYcMbnz7Jl0Lhxjscgxo2Dw4fhppvyOxxjjImoSFdcFzjHdMeh6oqa2rSBpk0jFpcxxkSCJQkfO3a4LjmOShI//ugGl7AKa2NMIWRJwseyZe7vUUnirbdcX01XXx2RmIwxJpIsSfg4pjuOAwdg4kTo0wfKlYtYXMYYEymWJHwkJUGVKlCtmjfjo49corCiJmNMIWVJwkdWdxxHBpx76y046yw4//yIxmWMMZFiScKTkQG//OJTH7FypevQ76abbJhSY0yhZUnCs24dpKT4JIkxY1xnfjfcENG4jDEmkixJeI7qjiMtDd59F7p396mgMMaYwseShCcpCYoUgUaNgM8/h+Rke8LaGFPoWZLwLFsG8fFQqhTuCeuaNeGyyyIdljHGRJQlCc+R7jg2b4bp02HAAFcnYYwxhZglCWD/fli/3ksS77wDmZlw442RDssYYyLOkgSu6StA86aZrlVTp05w+umRDcoYYwoASxJkt2w658As2LDBKqyNMcZjSQKXJMqWhWpfvA3ly8OVV0Y6JGOMKRAsSeBaNrVrtAv59FO47jqviZMxxphCnyRU3Z3EwOIfwF9/WVGTMcb4CGuSEJEuIrJaRNaKyP1B1ustIioiCd50PRFJEZEl3uv1cMW4aRPs3at02vg2nH02tGwZrkMZY0zUCduDACISB4wCLgU2A4tEZIqqrsixXlngTmBhjl2sU9WwX7Fr14atn/9E5W5L4cFXw304Y4yJKuG8k2gDrFXV9aqaBkwEevpZbwTwDJAaxlgCEoHqn78FJUtCv36RCMEYYwqscCaJmsAmn+nN3rwjRORsoLaqfuFn+/oi8rOIzBaRC/wdQEQGi0iiiCQmJyefWJSHDsH48XDVVa5lkzHGmCMiVnEtIkWA54G7/Sz+A6ijqq2AfwLjReSY8UNVdbSqJqhqQpUqVU4skL17oWtXGDz4xLY3xpgYFs7OibYAtX2ma3nzspQFmgLfiRvU5zRgioj0UNVE4C8AVV0sIuuAeCAxz6OsXh0mTMjz3RpjTCwI553EIqCBiNQXkeJAX2BK1kJV3auqlVW1nqrWA34AeqhqoohU8Sq+EZHTgQbA+jDGaowxxo+w3UmoarqIDANmAHHAGFVdLiLDgURVnRJk8wuB4SJyGMgEhqjqrnDFaowxxj9R1UjHkCcSEhI0MTHvS6OMMSaWichiVU0ItLzQP3FtjDEmMEsSxhhjArIkYYwxJiBLEsYYYwKyJGGMMSagmGndJCLJwG8nsYvKwI48CqcgiLXzgdg7p1g7H4i9c4q184Fjz6muqgbssiJmksTJEpHEYM3Aok2snQ/E3jnF2vlA7J1TrJ0PHP85WXGTMcaYgCxJGGOMCciSRLbRkQ4gj8Xa+UDsnVOsnQ/E3jnF2vnAcZ6T1UkYY4wJyO4kjDHGBGRJwhhjTECFPkmISBcRWS0ia0Xk/kjHkxdEZKOILBORJSISdV3jisgYEflTRH7xmVdRRL4WkV+9vxUiGePxCnBOj4nIFu97WiIiXSMZ4/EQkdoiMktEVojIchG505sfld9TkPOJ5u+opIj8KCJLvXN63JtfX0QWete8/3nj/QTeT2Guk/AGNloDXIobg3sR0E9VV0Q0sJMkIhuBBFWNyoeARORC4ADwnqo29eY9C+xS1ae9ZF5BVe+LZJzHI8A5PQYcUNXnIhnbiRCR6kB1Vf1JRMoCi4FewACi8HsKcj59iN7vSIAyqnpARIoBc4E7cUNCf6qqE0XkdWCpqr4WaD+F/U6iDbBWVderahowEegZ4ZgKPVX9Hsg5yFRP4F3v/bu4/8BRI8A5RS1V/UNVf/Le7wdWAjWJ0u8pyPlELXUOeJPFvJcCnYCPvfm5fkeFPUnUBDb5TG8myv9heBT4SkQWi8jgSAeTR6qp6h/e+21AtUgGk4eGiUiSVxwVFUUzOYlIPaAVsJAY+J5ynA9E8XckInEisgT4E/gaWAfsUdV0b5Vcr3mFPUnEqvaqejZwOTDUK+qIGerKSGOhnPQ14AygJfAH8N/IhnP8ROQU4BPgH6q6z3dZNH5Pfs4nqr8jVc1Q1ZZALVzJScPj3UdhTxJbgNo+07W8eVFNVbd4f/8EJuH+cUS77V65cVb58Z8Rjuekqep27z9xJvAmUfY9eeXcnwAfqOqn3uyo/Z78nU+0f0dZVHUPMAtoC5QXkaLeolyveYU9SSwCGni1/cWBvsCUCMd0UkSkjFfxhoiUAToDvwTfKipMAfp77/sDn0UwljyRdTH1XEkUfU9epejbwEpVfd5nUVR+T4HOJ8q/oyoiUt57XwrXQGclLllc5a2W63dUqFs3AXhN2kYCccAYVX0iwiGdFBE5HXf3AFAUGB9t5yQiE4COuC6NtwOPApOBD4E6uC7h+6hq1FQEBzinjrhiDAU2Arf4lOcXaCLSHpgDLAMyvdkP4srxo+57CnI+/Yje76g5rmI6DndD8KGqDveuEROBisDPwHWq+lfA/RT2JGGMMSawwl7cZIwxJghLEsYYYwKyJGGMMSYgSxLGGGMCsiRhjDEmIEsSxuRCRDJ8egFdkpe9BYtIPd+eYY0paIrmvooxhV6K17WBMYWO3UkYc4K8cTue9cbu+FFEzvTm1xORb71O4WaKSB1vfjURmeT1779URM73dhUnIm96ff5/5T0di4jc4Y1vkCQiEyN0mqaQsyRhTO5K5Shuutpn2V5VbQa8gntyH+Bl4F1VbQ58ALzkzX8JmK2qLYCzgeXe/AbAKFVtAuwBenvz7wdaefsZEq6TMyYYe+LamFyIyAFVPcXP/I1AJ1Vd73UOt01VK4nIDtwANoe9+X+oamURSQZq+XaB4HVL/bWqNvCm7wOKqer/ich03EBFk4HJPmMDGJNv7E7CmJOjAd4fD99+czLIriu8AhiFu+tY5NNzpzH5xpKEMSfnap+/C7z3UsWWrAAAAK5JREFU83E9CgNci+s4DmAmcCscGQzm1EA7FZEiQG1VnQXcB5wKHHM3Y0y42S8TY3JXyhvdK8t0Vc1qBltBRJJwdwP9vHm3A2NF5F4gGRjozb8TGC0iN+HuGG7FDWTjTxzwvpdIBHjJGxPAmHxldRLGnCCvTiJBVXdEOhZjwsWKm4wxxgRkdxLGGGMCsjsJY4wxAVmSMMYYE5AlCWOMMQFZkjDGGBOQJQljjDEB/T8mX/Diu2eQcwAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sg64DoycF60J",
        "colab_type": "text"
      },
      "source": [
        "Đối với validation accuracy, thuật toán ADAM cho độ chính xác cao hơn so với AMSGRAD trong phần lớn quá trình nhưng không ổn định. Và cũng như validation loss, độ chính xác của hai thuật toán này liên tục đổi ngôi cho nhau trong những lần lặp cuối."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YcJDU8_6GscU",
        "colab_type": "text"
      },
      "source": [
        "Kết luận:\n",
        "Trong bài báo 'On The Convergence Of ADAM And Beyond', các tác giả kết luận rằng AMSGRAD cho hiệu năng **tốt hơn nhiều** so với ADAM khi so sánh training loss và accuracy. Tuy nhiên, những kết quả của thí nghiệm ở trên lại không cho thấy điều tương tự. ADAM mang lại hiệu năng tương đương và tại một số điểm còn tốt hơn AMSGRAD."
      ]
    }
  ]
}