transformer-hv-demo.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-20T05:20:39.269124Z",
     "iopub.status.busy": "2021-10-20T05:20:39.26813Z",
     "iopub.status.idle": "2021-10-20T05:20:39.279016Z",
     "shell.execute_reply": "2021-10-20T05:20:39.278213Z",
     "shell.execute_reply.started": "2021-10-20T05:20:39.269072Z"
    }
   },
   "source": [
    "参考链接\n",
    "http://nlp.seas.harvard.edu/2018/04/03/attention.html </br>\n",
    "视频说明:\n",
    "https://www.bilibili.com/video/BV1pQ4y1z7Uf"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# transformer 实现英中翻译"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.003807Z",
     "iopub.status.busy": "2021-10-23T02:26:26.002897Z",
     "iopub.status.idle": "2021-10-23T02:26:26.014332Z",
     "shell.execute_reply": "2021-10-23T02:26:26.0136Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.003761Z"
    },
    "pycharm": {
     "is_executing": true
    }
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import jieba\n",
    "import seaborn\n",
    "import numpy as np\n",
    "import torch.nn as nn\n",
    "import math, copy, time\n",
    "import torch.nn.functional as F\n",
    "\n",
    "from tqdm import tqdm\n",
    "from torchtext import data\n",
    "from torch.nn import LayerNorm\n",
    "import matplotlib.pyplot as plt\n",
    "from torchtext.legacy.data import Field\n",
    "from torch.autograd import Variable\n",
    "from nltk.translate.bleu_score import sentence_bleu\n",
    "\n",
    "seaborn.set_context(context=\"talk\")\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.017547Z",
     "iopub.status.busy": "2021-10-23T02:26:26.017086Z",
     "iopub.status.idle": "2021-10-23T02:26:26.026748Z",
     "shell.execute_reply": "2021-10-23T02:26:26.02594Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.017302Z"
    }
   },
   "outputs": [],
   "source": [
    "class Config:\n",
    "    def __init__(self):\n",
    "        self.data_path = \"./data\"\n",
    "        self.testing_path = self.data_path + '/testing.txt'\n",
    "        self.training_path = self.data_path + '/training.txt'\n",
    "        self.validation_path = self.data_path + '/validation.txt'\n",
    "        self.init_token = '<bos>'#开始符号\n",
    "        self.eos_token = '<eos>'#终止符号\n",
    "        self.fix_length = 50\n",
    "        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
    "        #放到GPU上\n",
    "        self.batch_size = 128\n",
    "        #批次大小\n",
    "        self.beam_size = 4\n",
    "        self.n_layers = 6\n",
    "        self.h = 8\n",
    "        self.d_model = 512\n",
    "        self.d_ff = 2048\n",
    "        self.dropout = 0.1\n",
    "        self.epochs = 20\n",
    "        self.load_model = False               \n",
    "        self.load_model_path = 'weight.ckpt'     \n",
    "        self.smoothing = 0\n",
    "        self.factor = 2\n",
    "        self.warmup = 40\n",
    "        self.load_path = '../input/transformer-translate-entoch/transformer.pth'\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src='http://nlp.seas.harvard.edu/images/the-annotated-transformer_14_0.png'>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## embedding "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.02835Z",
     "iopub.status.busy": "2021-10-23T02:26:26.028097Z",
     "iopub.status.idle": "2021-10-23T02:26:26.039046Z",
     "shell.execute_reply": "2021-10-23T02:26:26.038303Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.028319Z"
    }
   },
   "outputs": [],
   "source": [
    "class Embeddings(nn.Module):\n",
    "    \"\"\"\n",
    "    d_model:512,Embeddings_size\n",
    "    \"\"\"\n",
    "    def __init__(self, d_model,  vocab_size):\n",
    "        super(Embeddings, self).__init__()\n",
    "        self.lut = nn.Embedding(vocab_size, d_model)\n",
    "        self.d_model = d_model\n",
    "\n",
    "    def forward(self, x):\n",
    "        return self.lut(x) * math.sqrt(self.d_model)#做了*开方的效果"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 位置编码"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "$$\n",
    "PE_{(pos,2i)} = sin(pos/10000^{2i/d_{model}}  \\quad) \\\\\n",
    "PE_{(pos,2i+1)} = cos(pos/10000^{2i/d_{model}}  \\quad)\n",
    "$$"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "$pos$表示是第几个词,$2i$和$2i+1$ 表示的是这个词第$2i$和$2i+1$个维度 \n",
    "维度是偶数用第一个公式，维度是奇数用第二个公式"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1, 100, 512)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#https://tensorflow.google.cn/tutorials/text/transformer\n",
    "import matplotlib.pyplot as plt\n",
    "import tensorflow as tf\n",
    "\n",
    "def get_angles(pos, i, d_model):\n",
    "  angle_rates = 1 / np.power(10000, (2 * (i//2)) / np.float32(d_model))\n",
    "  return pos * angle_rates\n",
    "\n",
    "def positional_encoding(position, d_model):\n",
    "  angle_rads = get_angles(np.arange(position)[:, np.newaxis],\n",
    "                          np.arange(d_model)[np.newaxis, :],\n",
    "                          d_model)\n",
    "\n",
    "  # 将 sin 应用于数组中的偶数索引（indices）；2i\n",
    "  angle_rads[:, 0::2] = np.sin(angle_rads[:, 0::2])\n",
    "\n",
    "  # 将 cos 应用于数组中的奇数索引；2i+1\n",
    "  angle_rads[:, 1::2] = np.cos(angle_rads[:, 1::2])\n",
    "\n",
    "  pos_encoding = angle_rads[np.newaxis, ...]\n",
    "\n",
    "  return tf.cast(pos_encoding, dtype=tf.float32)\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "pos_encoding = positional_encoding(100, 512)\n",
    "print (pos_encoding.shape)\n",
    "\n",
    "plt.pcolormesh(pos_encoding[0], cmap='RdBu')\n",
    "plt.xlabel('Depth')\n",
    "plt.xlim((0, 512))\n",
    "plt.ylabel('Position')\n",
    "plt.colorbar()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "但是会出现数值过大的问题，\n",
    "所以说我们用指数取代"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.040744Z",
     "iopub.status.busy": "2021-10-23T02:26:26.040426Z",
     "iopub.status.idle": "2021-10-23T02:26:26.051848Z",
     "shell.execute_reply": "2021-10-23T02:26:26.051192Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.040708Z"
    }
   },
   "outputs": [],
   "source": [
    "class PositionalEncoding(nn.Module):\n",
    "    \"Implement the PE function.\"\n",
    "    def __init__(self, d_model, dropout, max_len=5000):#d_model的默认值是512，每个词是512维\n",
    "        super(PositionalEncoding, self).__init__()\n",
    "        self.dropout = nn.Dropout(p=dropout)#dropout的几率\n",
    "        \n",
    "        # Compute the positional encodings once in log space.\n",
    "        pe = torch.zeros(max_len, d_model)#max_len*d_model维度的0矩阵\n",
    "        position = torch.arange(0, max_len).unsqueeze(1)#0,max_len-1的值按照维度1展开，（0是行，1是列）\n",
    "        #因为刚才的0矩阵的维度要匹配上位置\n",
    "        div_term = torch.exp(torch.arange(0, d_model, 2) *#0到d_model中，每隔2个的数值，如0,2,4,6,8...\n",
    "                             -(math.log(10000.0) / d_model))\n",
    "        pe[:, 0::2] = torch.sin(position * div_term)#维度跳跃，从0开始间隔为2\n",
    "        pe[:, 1::2] = torch.cos(position * div_term)#维度跳跃，从1开始间隔为2\n",
    "        pe = pe.unsqueeze(0)#解压到0维度\n",
    "        self.register_buffer('pe', pe)#加载到内存中，方便读取\n",
    "        \n",
    "    def forward(self, x):#前向传播\n",
    "        x = x + Variable(self.pe[:, :x.size(1)], \n",
    "                         requires_grad=False)\n",
    "        return self.dropout(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "https://tensorflow.google.cn/tutorials/text/transformer\n",
    "<img src='https://tensorflow.google.cn/tutorials/text/transformer_files/output_1kLCla68EloE_1.png'>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.055008Z",
     "iopub.status.busy": "2021-10-23T02:26:26.054695Z",
     "iopub.status.idle": "2021-10-23T02:26:26.320864Z",
     "shell.execute_reply": "2021-10-23T02:26:26.320097Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.054973Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "plt.figure(figsize=(15, 5))\n",
    "pe = PositionalEncoding(20,0)\n",
    "y = pe.forward(Variable(torch.zeros(1, 100, 20)))\n",
    "plt.plot(np.arange(100), y[0, :, 4:8].data.numpy())\n",
    "plt.legend([\"dim %d\"%p for p in [4,5,6,7]])\n",
    "None"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## selfAttention "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"https://gitee.com/xu_kai_xuyouqian/picture-bed/raw/master/20211013122948.png\"/>\n",
    "<img src=\"https://gitee.com/xu_kai_xuyouqian/picture-bed/raw/master/20211013123436.png\"/>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.322422Z",
     "iopub.status.busy": "2021-10-23T02:26:26.322068Z",
     "iopub.status.idle": "2021-10-23T02:26:26.330284Z",
     "shell.execute_reply": "2021-10-23T02:26:26.329578Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.322392Z"
    }
   },
   "outputs": [],
   "source": [
    "#注意力机制\n",
    "def attention(query, key, value, mask=None, dropout=None):\n",
    "    #q,k,v，\n",
    "    \"Compute 'Scaled Dot Product Attention'\"\n",
    "    d_k = query.size(-1)\n",
    "    # batch size, head size, query len , head dim\n",
    "    # batch size, head size, key len , head dim\n",
    "    scores = torch.matmul(query, key.transpose(-2, -1)) / math.sqrt(d_k)\n",
    "    if mask is not None:#掩码，这里是softmax指数\n",
    "        scores = scores.masked_fill(mask == 0, -1e9)\n",
    "    p_attn = F.softmax(scores, dim = -1)\n",
    "    if dropout is not None:\n",
    "        p_attn = dropout(p_attn)\n",
    "    return torch.matmul(p_attn, value), p_attn\n",
    "\n",
    "def clones(module, N):\n",
    "    \"Produce N identical layers.\"\n",
    "    return nn.ModuleList([copy.deepcopy(module) for _ in range(N)])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.333019Z",
     "iopub.status.busy": "2021-10-23T02:26:26.332711Z",
     "iopub.status.idle": "2021-10-23T02:26:26.345945Z",
     "shell.execute_reply": "2021-10-23T02:26:26.345263Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.332967Z"
    }
   },
   "outputs": [],
   "source": [
    "class MultiHeadedAttention(nn.Module):\n",
    "    def __init__(self, h, d_model, dropout=0.1):\n",
    "        \"Take in model size and number of heads.\"\n",
    "        super(MultiHeadedAttention, self).__init__()\n",
    "        assert d_model % h == 0\n",
    "        # We assume d_v always equals d_k\n",
    "        self.d_k = d_model // h\n",
    "        self.h = h#几个头\n",
    "        self.linears = clones(nn.Linear(d_model, d_model), 4)\n",
    "        #深复制4份\n",
    "        self.attn = None\n",
    "        self.dropout = nn.Dropout(p=dropout)\n",
    "        \n",
    "    def forward(self, query, key, value, mask=None):\n",
    "        \"\"\"\n",
    "        key len = value len\n",
    "        query = [batch size, query len, d_model]\n",
    "        key = [batch size, key len, d_model]\n",
    "        value = [batch size, value len, d_model]\n",
    "        mask = [batch size,1, key len] or [batch size, query len, key len]\n",
    "        \"\"\"\n",
    "        if mask is not None:\n",
    "            # Same mask applied to all h heads.\n",
    "            mask = mask.unsqueeze(1)\n",
    "        nbatches = query.size(0)\n",
    "        \n",
    "        # 1) Do all the linear projections in batch from d_model => h x d_k \n",
    "        query, key, value = \\\n",
    "            [l(x).view(nbatches, -1, self.h, self.d_k).transpose(1, 2)#view与reshape类似\n",
    "             for l, x in zip(self.linears, (query, key, value))]#这里是计算x与W_Q,W_K,W_V三个的产出QKV\n",
    "        \n",
    "        # 2) Apply attention on all the projected vectors in batch. \n",
    "        # x = batch size, h, query len, h_dim\n",
    "        x, self.attn = attention(query, key, value, mask=mask, \n",
    "                                 dropout=self.dropout)\n",
    "        \n",
    "        # 3) \"Concat\" using a view and apply a final linear. \n",
    "        x = x.transpose(1, 2).contiguous() \\\n",
    "             .view(nbatches, -1, self.h * self.d_k)\n",
    "        return self.linears[-1](x)#ffn"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## encoder"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### layernormal"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.348672Z",
     "iopub.status.busy": "2021-10-23T02:26:26.348343Z",
     "iopub.status.idle": "2021-10-23T02:26:26.36484Z",
     "shell.execute_reply": "2021-10-23T02:26:26.363868Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.348639Z"
    }
   },
   "outputs": [],
   "source": [
    "# layer normal 演示\n",
    "import torch\n",
    "from torch import nn\n",
    "output = torch.range(1, 24, dtype=torch.float32).contiguous().view(2, 3, 4)\n",
    "output[0][0][0] = 2\n",
    "output[-1][-1][-1] = 30\n",
    "print('*' * 20 + '原始数据' + '*' * 20)\n",
    "print(output)\n",
    "print('*' * 20 + 'layernormal结果' + '*' * 20)\n",
    "layernormal = nn.LayerNorm(4)(output)\n",
    "print(layernormal)\n",
    "mean = torch.mean(output, dim=2).unsqueeze(2)\n",
    "std = torch.std(output, dim=2).unsqueeze(2)\n",
    "print('*' * 20 + '手动layernormal结果' + '*' * 20)\n",
    "print((output - mean) / std * (2 / 3 ** (0.5)))\n",
    "'''\n",
    "\n",
    "********************原始数据********************\n",
    "tensor([[[ 2.,  2.,  3.,  4.],\n",
    "         [ 5.,  6.,  7.,  8.],\n",
    "         [ 9., 10., 11., 12.]],\n",
    "\n",
    "        [[13., 14., 15., 16.],\n",
    "         [17., 18., 19., 20.],\n",
    "         [21., 22., 23., 30.]]])\n",
    "********************layernormal结果********************\n",
    "tensor([[[-0.9045, -0.9045,  0.3015,  1.5075],\n",
    "         [-1.3416, -0.4472,  0.4472,  1.3416],\n",
    "         [-1.3416, -0.4472,  0.4472,  1.3416]],\n",
    "\n",
    "        [[-1.3416, -0.4472,  0.4472,  1.3416],\n",
    "         [-1.3416, -0.4472,  0.4472,  1.3416],\n",
    "         [-0.8485, -0.5657, -0.2828,  1.6971]]],\n",
    "       grad_fn=<NativeLayerNormBackward>)\n",
    "********************手动layernormal结果********************\n",
    "tensor([[[-0.9045, -0.9045,  0.3015,  1.5076],\n",
    "         [-1.3416, -0.4472,  0.4472,  1.3416],\n",
    "         [-1.3416, -0.4472,  0.4472,  1.3416]],\n",
    "\n",
    "        [[-1.3416, -0.4472,  0.4472,  1.3416],\n",
    "         [-1.3416, -0.4472,  0.4472,  1.3416],\n",
    "         [-0.8485, -0.5657, -0.2828,  1.6971]]])\n",
    "\n",
    "\n",
    "'''\n",
    "None"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.36671Z",
     "iopub.status.busy": "2021-10-23T02:26:26.366302Z",
     "iopub.status.idle": "2021-10-23T02:26:26.374962Z",
     "shell.execute_reply": "2021-10-23T02:26:26.374039Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.366662Z"
    }
   },
   "outputs": [],
   "source": [
    "class PositionwiseFeedForward(nn.Module):\n",
    "    \"Implements FFN equation.\"\n",
    "    def __init__(self, d_model, d_ff, dropout=0.1):\n",
    "        super(PositionwiseFeedForward, self).__init__()\n",
    "        self.w_1 = nn.Linear(d_model, d_ff)\n",
    "        self.w_2 = nn.Linear(d_ff, d_model)\n",
    "        self.dropout = nn.Dropout(dropout)\n",
    "\n",
    "    def forward(self, x):\n",
    "        return self.w_2(self.dropout(F.relu(self.w_1(x))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.376919Z",
     "iopub.status.busy": "2021-10-23T02:26:26.376591Z",
     "iopub.status.idle": "2021-10-23T02:26:26.385233Z",
     "shell.execute_reply": "2021-10-23T02:26:26.384519Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.376884Z"
    }
   },
   "outputs": [],
   "source": [
    "class SublayerConnection(nn.Module):\n",
    "    \"\"\"\n",
    "    A residual connection followed by a layer norm.\n",
    "    Note for code simplicity the norm is first as opposed to last.\n",
    "\n",
    "    add&norm\n",
    "    \"\"\"\n",
    "    def __init__(self, size, dropout):\n",
    "        super(SublayerConnection, self).__init__()\n",
    "        self.norm = LayerNorm(size)\n",
    "        self.dropout = nn.Dropout(dropout)\n",
    "\n",
    "    def forward(self, x, sublayer):\n",
    "        \"Apply residual connection to any sublayer with the same size.\"\n",
    "        return x + self.dropout(sublayer(self.norm(x)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.38739Z",
     "iopub.status.busy": "2021-10-23T02:26:26.386391Z",
     "iopub.status.idle": "2021-10-23T02:26:26.396338Z",
     "shell.execute_reply": "2021-10-23T02:26:26.395559Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.387355Z"
    }
   },
   "outputs": [],
   "source": [
    "class EncoderLayer(nn.Module):\n",
    "    \"Encoder is made up of self-attn and feed forward (defined below)\"\n",
    "    def __init__(self, size, self_attn, feed_forward, dropout):\n",
    "        super(EncoderLayer, self).__init__()\n",
    "        self.self_attn = self_attn\n",
    "        self.feed_forward = feed_forward\n",
    "        self.sublayer = clones(SublayerConnection(size, dropout), 2)\n",
    "        self.size = size\n",
    "\n",
    "    def forward(self, x, mask):\n",
    "        \"Follow Figure 1 (left) for connections.\"\n",
    "        x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, mask))\n",
    "        return self.sublayer[1](x, self.feed_forward)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.400963Z",
     "iopub.status.busy": "2021-10-23T02:26:26.400716Z",
     "iopub.status.idle": "2021-10-23T02:26:26.408311Z",
     "shell.execute_reply": "2021-10-23T02:26:26.407398Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.400914Z"
    }
   },
   "outputs": [],
   "source": [
    "class Encoder(nn.Module):\n",
    "    \"Core encoder is a stack of N layers\"\n",
    "    def __init__(self, layer, N):\n",
    "        super(Encoder, self).__init__()\n",
    "        self.layers = clones(layer, N)\n",
    "        self.norm = LayerNorm(layer.size)\n",
    "        \n",
    "    def forward(self, x, mask):\n",
    "        \"Pass the input (and mask) through each layer in turn.\"\n",
    "        for layer in self.layers:\n",
    "            x = layer(x, mask)\n",
    "        return self.norm(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## decoder"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### decoder mask"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.409927Z",
     "iopub.status.busy": "2021-10-23T02:26:26.409537Z",
     "iopub.status.idle": "2021-10-23T02:26:26.419965Z",
     "shell.execute_reply": "2021-10-23T02:26:26.419143Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.409892Z"
    }
   },
   "outputs": [],
   "source": [
    "def subsequent_mask(size):\n",
    "    \"Mask out subsequent positions.\"\n",
    "    #下三角矩阵\n",
    "    attn_shape = (1, size, size)\n",
    "    subsequent_mask = np.triu(np.ones(attn_shape), k=1).astype('uint8')\n",
    "    #上三角矩阵，且往上移动一下\n",
    "    \"\"\"\n",
    "        [\n",
    "        [0(k=0),0(k=1),1]\n",
    "        [0,0(k=0),0(k=1)]\n",
    "        [0,0,0(k=0)]\n",
    "        ]\n",
    "    \"\"\"\n",
    "    return torch.from_numpy(subsequent_mask) == 0\n",
    "print(subsequent_mask(4))\n",
    "\"\"\"\n",
    "tensor([[[ True, False, False, False],\n",
    "         [ True,  True, False, False],\n",
    "         [ True,  True,  True, False],\n",
    "         [ True,  True,  True,  True]]])\n",
    "\"\"\"\n",
    "None"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.426962Z",
     "iopub.status.busy": "2021-10-23T02:26:26.426705Z",
     "iopub.status.idle": "2021-10-23T02:26:26.43731Z",
     "shell.execute_reply": "2021-10-23T02:26:26.436617Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.426933Z"
    }
   },
   "outputs": [],
   "source": [
    "class DecoderLayer(nn.Module):\n",
    "    \"Decoder is made of self-attn, src-attn, and feed forward (defined below)\"\n",
    "    def __init__(self, size, self_attn, src_attn, feed_forward, dropout):\n",
    "        super(DecoderLayer, self).__init__()\n",
    "        self.size = size\n",
    "        self.self_attn = self_attn\n",
    "        self.src_attn = src_attn\n",
    "        self.feed_forward = feed_forward\n",
    "        self.sublayer = clones(SublayerConnection(size, dropout), 3)\n",
    " \n",
    "    def forward(self, x, memory, src_mask, tgt_mask):\n",
    "        \"\"\"\n",
    "        x: decoder input\n",
    "        memory: encoder output\n",
    "        \"\"\"\n",
    "        m = memory\n",
    "        x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, tgt_mask))\n",
    "        x = self.sublayer[1](x, lambda x: self.src_attn(x, m, m, src_mask))\n",
    "        return self.sublayer[2](x, self.feed_forward)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.440704Z",
     "iopub.status.busy": "2021-10-23T02:26:26.440399Z",
     "iopub.status.idle": "2021-10-23T02:26:26.447763Z",
     "shell.execute_reply": "2021-10-23T02:26:26.447036Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.440648Z"
    }
   },
   "outputs": [],
   "source": [
    "class Decoder(nn.Module):\n",
    "    \"Generic N layer decoder with masking.\"\n",
    "    def __init__(self, layer, N):\n",
    "        super(Decoder, self).__init__()\n",
    "        self.layers = clones(layer, N)\n",
    "        self.norm = LayerNorm(layer.size)\n",
    "        \n",
    "    def forward(self, x, memory, src_mask, tgt_mask):\n",
    "        for layer in self.layers:\n",
    "            x = layer(x, memory, src_mask, tgt_mask)\n",
    "        return self.norm(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 全连接"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.449428Z",
     "iopub.status.busy": "2021-10-23T02:26:26.449031Z",
     "iopub.status.idle": "2021-10-23T02:26:26.456434Z",
     "shell.execute_reply": "2021-10-23T02:26:26.455672Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.449393Z"
    }
   },
   "outputs": [],
   "source": [
    "class Generator(nn.Module):\n",
    "    \"Define standard linear + softmax generation step.\"\n",
    "    def __init__(self, d_model, vocab):\n",
    "        super(Generator, self).__init__()\n",
    "        self.proj = nn.Linear(d_model, vocab)\n",
    "\n",
    "    def forward(self, x):\n",
    "        return F.log_softmax(self.proj(x), dim=-1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 完整模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.458361Z",
     "iopub.status.busy": "2021-10-23T02:26:26.458014Z",
     "iopub.status.idle": "2021-10-23T02:26:26.467072Z",
     "shell.execute_reply": "2021-10-23T02:26:26.466213Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.458272Z"
    }
   },
   "outputs": [],
   "source": [
    "class EncoderDecoder(nn.Module):\n",
    "    \"\"\"\n",
    "    A standard Encoder-Decoder architecture. Base for this and many \n",
    "    other models.\n",
    "    \"\"\"\n",
    "    def __init__(self, encoder, decoder, src_embed, tgt_embed, generator):\n",
    "        super(EncoderDecoder, self).__init__()\n",
    "        self.encoder = encoder\n",
    "        self.decoder = decoder\n",
    "        self.src_embed = src_embed\n",
    "        self.tgt_embed = tgt_embed\n",
    "        self.generator = generator\n",
    "        \n",
    "    def forward(self, src, tgt, src_mask, tgt_mask):\n",
    "        \"Take in and process masked src and target sequences.\"\n",
    "        return self.decode(self.encode(src, src_mask), src_mask,\n",
    "                            tgt, tgt_mask)\n",
    "    \n",
    "    def encode(self, src, src_mask):\n",
    "        return self.encoder(self.src_embed(src), src_mask)\n",
    "    \n",
    "    def decode(self, memory, src_mask, tgt, tgt_mask):\n",
    "        return self.decoder(self.tgt_embed(tgt), memory, src_mask, tgt_mask)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-20T07:21:10.373845Z",
     "iopub.status.busy": "2021-10-20T07:21:10.373479Z",
     "iopub.status.idle": "2021-10-20T07:21:10.381429Z",
     "shell.execute_reply": "2021-10-20T07:21:10.380757Z",
     "shell.execute_reply.started": "2021-10-20T07:21:10.37381Z"
    }
   },
   "source": [
    "## 创建模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.468962Z",
     "iopub.status.busy": "2021-10-23T02:26:26.468711Z",
     "iopub.status.idle": "2021-10-23T02:26:26.478576Z",
     "shell.execute_reply": "2021-10-23T02:26:26.477878Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.468931Z"
    }
   },
   "outputs": [],
   "source": [
    "def make_model(src_vocab, tgt_vocab, N=6, \n",
    "               d_model=512, d_ff=2048, h=8, dropout=0.1):\n",
    "    \"Helper: Construct a model from hyperparameters.\"\n",
    "    c = copy.deepcopy\n",
    "    attn = MultiHeadedAttention(h, d_model)\n",
    "    ff = PositionwiseFeedForward(d_model, d_ff, dropout)\n",
    "    position = PositionalEncoding(d_model, dropout)\n",
    "    model = EncoderDecoder(\n",
    "        Encoder(EncoderLayer(d_model, c(attn), c(ff), dropout), N),\n",
    "        Decoder(DecoderLayer(d_model, c(attn), c(attn), c(ff), dropout), N),\n",
    "        nn.Sequential(Embeddings(d_model, src_vocab), c(position)),\n",
    "        nn.Sequential(Embeddings(d_model, tgt_vocab), c(position)),\n",
    "        Generator(d_model, tgt_vocab))\n",
    "    \n",
    "    # This was important from their code. \n",
    "    # Initialize parameters with Glorot / fan_avg.\n",
    "    for p in model.parameters():\n",
    "        if p.dim() > 1:\n",
    "            nn.init.xavier_uniform(p)\n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-20T07:21:10.395323Z",
     "iopub.status.busy": "2021-10-20T07:21:10.395086Z",
     "iopub.status.idle": "2021-10-20T07:21:10.403419Z",
     "shell.execute_reply": "2021-10-20T07:21:10.402668Z",
     "shell.execute_reply.started": "2021-10-20T07:21:10.395289Z"
    }
   },
   "source": [
    "## batch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.480254Z",
     "iopub.status.busy": "2021-10-23T02:26:26.479979Z",
     "iopub.status.idle": "2021-10-23T02:26:26.489139Z",
     "shell.execute_reply": "2021-10-23T02:26:26.488184Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.480219Z"
    }
   },
   "outputs": [],
   "source": [
    "class Batch:\n",
    "    \"Object for holding a batch of data with mask during training.\"\n",
    "    def __init__(self, src, trg=None, pad=0):\n",
    "        \"\"\"\n",
    "        src = batch size, seq len\n",
    "        \n",
    "        \"\"\"\n",
    "        self.src = src\n",
    "        # src_mask = batch size, 1, seq len\n",
    "        self.src_mask = (src != pad).unsqueeze(-2)\n",
    "        if trg is not None:\n",
    "            self.trg = trg[:, :-1]\n",
    "            self.trg_y = trg[:, 1:]\n",
    "            self.trg_mask = \\\n",
    "                self.make_std_mask(self.trg, pad)\n",
    "            self.ntokens = (self.trg_y != pad).data.sum()\n",
    "    \n",
    "    @staticmethod\n",
    "    def make_std_mask(tgt, pad):\n",
    "        \"Create a mask to hide padding and future words.\"\n",
    "        tgt_mask = (tgt != pad).unsqueeze(-2)\n",
    "        tgt_mask = tgt_mask & Variable(\n",
    "            subsequent_mask(tgt.size(-1)).type_as(tgt_mask.data))\n",
    "        return tgt_mask"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-20T07:21:10.418983Z",
     "iopub.status.busy": "2021-10-20T07:21:10.418405Z",
     "iopub.status.idle": "2021-10-20T07:21:10.427169Z",
     "shell.execute_reply": "2021-10-20T07:21:10.426444Z",
     "shell.execute_reply.started": "2021-10-20T07:21:10.418913Z"
    }
   },
   "source": [
    "## warm up"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.491067Z",
     "iopub.status.busy": "2021-10-23T02:26:26.490772Z",
     "iopub.status.idle": "2021-10-23T02:26:26.502017Z",
     "shell.execute_reply": "2021-10-23T02:26:26.501303Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.490996Z"
    }
   },
   "outputs": [],
   "source": [
    "#warmup\n",
    "class NoamOpt:\n",
    "    \"Optim wrapper that implements rate.\"\n",
    "    def __init__(self, model_size, factor, warmup, optimizer):\n",
    "        self.optimizer = optimizer\n",
    "        self._step = 0\n",
    "        self.warmup = warmup\n",
    "        self.factor = factor\n",
    "        self.model_size = model_size\n",
    "        self._rate = 0\n",
    "\n",
    "    def zero_grad(self):\n",
    "        self.optimizer.zero_grad()\n",
    "    def step(self):\n",
    "        \"Update parameters and rate\"\n",
    "        self._step += 1\n",
    "        rate = self.rate()\n",
    "        for p in self.optimizer.param_groups:\n",
    "            p['lr'] = rate\n",
    "        self._rate = rate\n",
    "        self.optimizer.step()\n",
    "        \n",
    "    def rate(self, step = None):\n",
    "        \"Implement `lrate` above\"\n",
    "        if step is None:\n",
    "            step = self._step\n",
    "        return self.factor * \\\n",
    "            (self.model_size ** (-0.5) *\n",
    "            min(step ** (-0.5), step * self.warmup ** (-1.5)))\n",
    "        \n",
    "def get_std_opt(model):\n",
    "    return NoamOpt(model.src_embed[0].d_model, 2, 4000,\n",
    "            torch.optim.Adam(model.parameters(), lr=0, betas=(0.9, 0.98), eps=1e-9))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:26.503726Z",
     "iopub.status.busy": "2021-10-23T02:26:26.503314Z",
     "iopub.status.idle": "2021-10-23T02:26:27.059683Z",
     "shell.execute_reply": "2021-10-23T02:26:27.058986Z",
     "shell.execute_reply.started": "2021-10-23T02:26:26.503693Z"
    }
   },
   "outputs": [],
   "source": [
    "# Three settings of the lrate hyperparameters.\n",
    "opts = [NoamOpt(512, 1, 4000, None), \n",
    "        NoamOpt(512, 1, 8000, None),\n",
    "        NoamOpt(256, 1, 4000, None)]\n",
    "plt.plot(np.arange(1, 20000), [[opt.rate(i) for opt in opts] for i in range(1, 20000)])\n",
    "plt.legend([\"512:4000\", \"512:8000\", \"256:4000\"])\n",
    "None"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-20T07:21:10.451682Z",
     "iopub.status.busy": "2021-10-20T07:21:10.451401Z",
     "iopub.status.idle": "2021-10-20T07:21:10.459339Z",
     "shell.execute_reply": "2021-10-20T07:21:10.458314Z",
     "shell.execute_reply.started": "2021-10-20T07:21:10.451648Z"
    }
   },
   "source": [
    "## 标签平滑"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-20T07:21:10.461021Z",
     "iopub.status.busy": "2021-10-20T07:21:10.460555Z",
     "iopub.status.idle": "2021-10-20T07:21:10.468377Z",
     "shell.execute_reply": "2021-10-20T07:21:10.467441Z",
     "shell.execute_reply.started": "2021-10-20T07:21:10.460986Z"
    },
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "source": [
    "#文本翻译更看重这个\n",
    "#kl散度\n",
    "\"\"\" python\n",
    "预测softmax_res = [0.1, 0.3, 0.6]\n",
    "真实结果是[2] -->[0,0,1]\n",
    "如果不做平滑直接是这两者计算交叉信息熵\n",
    "如果做了平滑可以把真实结果修改为\n",
    "[0.1,0.1,0.8]\n",
    "计算他们的kl散度\"\"\"\n"
   ],
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:27.061304Z",
     "iopub.status.busy": "2021-10-23T02:26:27.061038Z",
     "iopub.status.idle": "2021-10-23T02:26:27.069447Z",
     "shell.execute_reply": "2021-10-23T02:26:27.068806Z",
     "shell.execute_reply.started": "2021-10-23T02:26:27.061272Z"
    }
   },
   "outputs": [],
   "source": [
    "class LabelSmoothing(nn.Module):\n",
    "    \"Implement label smoothing.\"\n",
    "    \"\"\"\n",
    "    size:预测词典长度\n",
    "    smoothing:分给真实标签，1-smoothing，其余的smoothing/(n-1)\n",
    "    \"\"\"\n",
    "    def __init__(self, size, padding_idx, smoothing=0.0):\n",
    "        super(LabelSmoothing, self).__init__()\n",
    "        self.criterion = nn.KLDivLoss(size_average=False)\n",
    "        self.padding_idx = padding_idx\n",
    "        self.confidence = 1.0 - smoothing\n",
    "        self.smoothing = smoothing\n",
    "        self.size = size\n",
    "        self.true_dist = None\n",
    "        \n",
    "    def forward(self, x, target):\n",
    "        assert x.size(1) == self.size\n",
    "        true_dist = x.data.clone()\n",
    "        true_dist.fill_(self.smoothing / (self.size - 2))\n",
    "        true_dist.scatter_(1, target.data.unsqueeze(1), self.confidence)\n",
    "        true_dist[:, self.padding_idx] = 0\n",
    "        mask = torch.nonzero(target.data == self.padding_idx)\n",
    "        if mask.dim() > 0:\n",
    "            true_dist.index_fill_(0, mask.squeeze(), 0.0)\n",
    "        self.true_dist = true_dist\n",
    "        return self.criterion(x, Variable(true_dist, requires_grad=False))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 计算损失"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:27.071233Z",
     "iopub.status.busy": "2021-10-23T02:26:27.070786Z",
     "iopub.status.idle": "2021-10-23T02:26:27.07927Z",
     "shell.execute_reply": "2021-10-23T02:26:27.078589Z",
     "shell.execute_reply.started": "2021-10-23T02:26:27.071201Z"
    }
   },
   "outputs": [],
   "source": [
    "class SimpleLossCompute:\n",
    "    \"A simple loss compute and train function.\"\n",
    "    def __init__(self, generator, criterion, opt=None):\n",
    "        self.generator = generator\n",
    "        self.criterion = criterion\n",
    "        self.opt = opt\n",
    "        \n",
    "    def __call__(self, x, y, norm):\n",
    "        \"\"\"\n",
    "        norm 非pad的单词个数\n",
    "        \"\"\"\n",
    "        x = self.generator(x)\n",
    "        loss = self.criterion(x.contiguous().view(-1, x.size(-1)), y.contiguous().view(-1)) / norm\n",
    "        loss.backward()\n",
    "        if self.opt is not None:\n",
    "            self.opt.step()\n",
    "            self.opt.optimizer.zero_grad()\n",
    "        return loss.item() * norm"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-20T07:21:10.490676Z",
     "iopub.status.busy": "2021-10-20T07:21:10.490415Z",
     "iopub.status.idle": "2021-10-20T07:21:10.499847Z",
     "shell.execute_reply": "2021-10-20T07:21:10.498855Z",
     "shell.execute_reply.started": "2021-10-20T07:21:10.490645Z"
    }
   },
   "source": [
    "## 训练单个epoch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:27.080924Z",
     "iopub.status.busy": "2021-10-23T02:26:27.080646Z",
     "iopub.status.idle": "2021-10-23T02:26:27.089357Z",
     "shell.execute_reply": "2021-10-23T02:26:27.088686Z",
     "shell.execute_reply.started": "2021-10-23T02:26:27.080891Z"
    }
   },
   "outputs": [],
   "source": [
    "def run_epoch(data_iter, model, loss_compute):\n",
    "    \"Standard Training and Logging Function\"\n",
    "    start = time.time()\n",
    "    total_tokens = 0\n",
    "    total_loss = 0\n",
    "    tokens = 0\n",
    "    for i, batch in enumerate(data_iter):\n",
    "        out = model.forward(batch.src, batch.trg, batch.src_mask, batch.trg_mask)\n",
    "        loss = loss_compute(out, batch.trg_y, batch.ntokens)\n",
    "        total_loss += loss\n",
    "        total_tokens += batch.ntokens\n",
    "        tokens += batch.ntokens\n",
    "        if i % 50 == 1:\n",
    "            elapsed = time.time() - start\n",
    "            print(\"Epoch Step: %d Loss: %f Tokens per Sec: %f\" %\n",
    "                    (i, loss / batch.ntokens, tokens / elapsed))\n",
    "            start = time.time()\n",
    "            tokens = 0\n",
    "    return total_loss / total_tokens"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 模拟训练"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:27.091032Z",
     "iopub.status.busy": "2021-10-23T02:26:27.090772Z",
     "iopub.status.idle": "2021-10-23T02:26:27.101215Z",
     "shell.execute_reply": "2021-10-23T02:26:27.100226Z",
     "shell.execute_reply.started": "2021-10-23T02:26:27.090998Z"
    }
   },
   "outputs": [],
   "source": [
    "def data_gen(V, batch, nbatches):\n",
    "    \"Generate random data for a src-tgt copy task.\"\n",
    "    \"\"\"\n",
    "    V : 词汇表长度\n",
    "    nbatches : 有多少个batch\n",
    "    batch：batch size\n",
    "    \"\"\"\n",
    "    for i in range(nbatches):\n",
    "        # 10 是句子长度\n",
    "        data = torch.from_numpy(np.random.randint(1, V, size=(batch, 10)))\n",
    "        data[:, 0] = 1\n",
    "        src = Variable(data, requires_grad=False)\n",
    "        tgt = Variable(data, requires_grad=False)\n",
    "        yield Batch(src, tgt, 0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:26:27.102776Z",
     "iopub.status.busy": "2021-10-23T02:26:27.102496Z",
     "iopub.status.idle": "2021-10-23T02:27:54.479005Z",
     "shell.execute_reply": "2021-10-23T02:27:54.478113Z",
     "shell.execute_reply.started": "2021-10-23T02:26:27.102742Z"
    }
   },
   "outputs": [],
   "source": [
    "V = 11  # vocabulary size\n",
    "criterion = LabelSmoothing(size=V, padding_idx=0, smoothing=0.0)\n",
    "model = make_model(V, V, N=2)\n",
    "model_opt = NoamOpt(model.src_embed[0].d_model, 1, 400,\n",
    "        torch.optim.Adam(model.parameters(), lr=0, betas=(0.9, 0.98), eps=1e-9))\n",
    "\n",
    "for epoch in range(10):\n",
    "    model.train()\n",
    "    run_epoch(data_gen(V, 30, 20), model, \n",
    "              SimpleLossCompute(model.generator, criterion, model_opt))\n",
    "    model.eval()\n",
    "    print(run_epoch(data_gen(V, 30, 5), model, \n",
    "                    SimpleLossCompute(model.generator, criterion, None)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 贪心解码"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.483197Z",
     "iopub.status.busy": "2021-10-23T02:27:54.482861Z",
     "iopub.status.idle": "2021-10-23T02:27:54.605353Z",
     "shell.execute_reply": "2021-10-23T02:27:54.604601Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.483169Z"
    }
   },
   "outputs": [],
   "source": [
    "def greedy_decode(model, src, src_mask, max_len, start_symbol):\n",
    "    memory = model.encode(src, src_mask)\n",
    "    batch_size, _ = src.shape\n",
    "    ys = torch.ones(batch_size, 1).fill_(start_symbol).type_as(src.data)\n",
    "\n",
    "    for i in range(max_len-1):\n",
    "        out = model.decode(memory, src_mask, \n",
    "                           Variable(ys), \n",
    "                           Variable(subsequent_mask(ys.size(1))\n",
    "                                    .type_as(src.data)))\n",
    "        \n",
    "        # prob = batch size, vocab size\n",
    "        prob = model.generator(out[:, -1])\n",
    "        _, next_word = torch.max(prob, dim = -1)\n",
    "        \n",
    "        ys = torch.cat([ys, next_word.unsqueeze(1)], dim=-1)\n",
    "    return ys\n",
    "\n",
    "src = Variable(torch.LongTensor([[1,2,3,4,5,6,7,8,9,10],\n",
    "                                 [1,2,3,4,8,6,7,8,9,10]\n",
    "                                ]) )\n",
    "src_mask = Variable(torch.ones(2, 1, 10) )\n",
    "print(src_mask)\n",
    "print(greedy_decode(model, src, src_mask, max_len=10, start_symbol=1)[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## beam search 解码"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.607655Z",
     "iopub.status.busy": "2021-10-23T02:27:54.606564Z",
     "iopub.status.idle": "2021-10-23T02:27:54.616064Z",
     "shell.execute_reply": "2021-10-23T02:27:54.615151Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.607615Z"
    }
   },
   "outputs": [],
   "source": [
    "class Hypothesis:\n",
    "    \"\"\"\n",
    "    保存beamsearch 的备选方案\n",
    "    \"\"\"\n",
    "\n",
    "    def __init__(self, tokens, log_probs):\n",
    "        self.tokens = tokens  # list of all the tokens from time 0 to the current time step t\n",
    "        self.log_probs = log_probs  # list of the log probabilities of the tokens of the tokens\n",
    "\n",
    "    def extend(self, token, log_prob):\n",
    "        \"\"\"\n",
    "        Method to extend the current hypothesis by adding the next decoded token and all the informations associated with it\n",
    "        \"\"\"\n",
    "        return Hypothesis(tokens=torch.cat([self.tokens, token]),  # we add the decoded token\n",
    "                          log_probs=self.log_probs + [log_prob],  # we add the log prob of the decoded token\n",
    "                          )\n",
    "\n",
    "    @property\n",
    "    def latest_token(self):\n",
    "        return self.tokens[-1]\n",
    "\n",
    "    @property\n",
    "    def tot_log_prob(self):\n",
    "        return sum(self.log_probs)\n",
    "\n",
    "    @property\n",
    "    def avg_log_prob(self):\n",
    "        return self.tot_log_prob / len(self.tokens)\n",
    "\n",
    "def get_top_k_beam(prediction, k):\n",
    "    top_k_probs, top_k_ids = nn.functional.softmax(prediction, 0).topk(k)\n",
    "\n",
    "    top_k_log_probs = torch.log(top_k_probs)\n",
    "    return top_k_log_probs, top_k_ids"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.617671Z",
     "iopub.status.busy": "2021-10-23T02:27:54.61735Z",
     "iopub.status.idle": "2021-10-23T02:27:54.63314Z",
     "shell.execute_reply": "2021-10-23T02:27:54.632263Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.617638Z"
    }
   },
   "outputs": [],
   "source": [
    "def beam_search(model, src, src_mask, max_len, start_symbol, beam_size, device):\n",
    "    # memory = batch size, src len, d_model\n",
    "    memory = model.encode(src, src_mask)\n",
    "    batch_size, _ = src.shape\n",
    "    all_tokens = []\n",
    "    for batch in tqdm(range(batch_size)):\n",
    "        batch_memory = memory[batch]\n",
    "        batch_src_mask = src_mask[batch]\n",
    "\n",
    "        tokens = beam_search_single_batch(model, batch_memory, batch_src_mask,\n",
    "                                                  beam_size, start_symbol, max_len, device)\n",
    "        \n",
    "        all_tokens.append(tokens)\n",
    "    all_tokens = torch.stack(all_tokens)\n",
    "    return all_tokens"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.635112Z",
     "iopub.status.busy": "2021-10-23T02:27:54.634714Z",
     "iopub.status.idle": "2021-10-23T02:27:54.647619Z",
     "shell.execute_reply": "2021-10-23T02:27:54.64653Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.635072Z"
    }
   },
   "outputs": [],
   "source": [
    "def beam_search_single_batch(model, memory, src_mask, beam_size, start_symbol, max_len, device):\n",
    "    \"\"\"\n",
    "    memory = [src len, d_model]\n",
    "    \"\"\"\n",
    "    # memory = [beam size, src len, d_model]\n",
    "    memory = memory.unsqueeze(0).repeat(beam_size, 1, 1)\n",
    "    hys = [Hypothesis(torch.LongTensor([start_symbol]).to(device), [0]) for _ in range(beam_size)]\n",
    "    \n",
    "    for i in range(max_len-1):\n",
    "        # beam_input = [beam size, decoded len]\n",
    "        beam_input = torch.stack([h.tokens for h in hys])\n",
    "        trg_mask= subsequent_mask(beam_input.size(1)).type_as(memory.data)\n",
    "        \n",
    "        out = model.decode(memory, src_mask, beam_input, trg_mask)\n",
    "        probs = model.generator(out[:,-1])  # beam size, vocab size\n",
    "        \n",
    "        # 把之前的最优解再扩展一个词，先要获取之前最优解的个数\n",
    "        num_ori_hyps = 1 if i == 0 else len(hys)\n",
    " \n",
    "        # 创建列表保存新的结果\n",
    "        all_hyps = []\n",
    "        \n",
    "        # 遍历添加所有可能结果\n",
    "        for j in range(num_ori_hyps):\n",
    "            prob = probs[j, :]\n",
    "            top_k_log_probs, top_k_ids = get_top_k_beam(prob, beam_size)\n",
    "            \n",
    "            for log_prob, top_k_id in zip(top_k_log_probs, top_k_ids):\n",
    "                top_k_id = top_k_id.view(-1)\n",
    "                new_hys = hys[j].extend(top_k_id, log_prob)\n",
    "                all_hyps.append(new_hys)\n",
    "        \n",
    "        \n",
    "        # 重置\n",
    "        hys = []\n",
    "        # 按照概率来排序\n",
    "        sorted_hyps = sorted(all_hyps, key = lambda h:h.avg_log_prob, reverse=True)\n",
    "        hys = sorted_hyps[:beam_size]\n",
    "    \n",
    "    tokens = hys[0].tokens\n",
    "    return tokens\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 数据准备"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.649793Z",
     "iopub.status.busy": "2021-10-23T02:27:54.649237Z",
     "iopub.status.idle": "2021-10-23T02:27:54.667167Z",
     "shell.execute_reply": "2021-10-23T02:27:54.666381Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.649738Z"
    }
   },
   "outputs": [],
   "source": [
    "class MyDataset(data.Dataset):\n",
    "\n",
    "    def __init__(self, datatuple, text_field, label_field, test=False):  # datatuple指的是元组('this moive is great',1)\n",
    "        fields = [(\"text\", text_field), (\"label\", label_field)]\n",
    "        lists = []\n",
    "        for content, label in tqdm(datatuple):\n",
    "            # Example: Defines a single training or test example.Stores each column of the example as an attribute.\n",
    "            lists.append(data.Example.fromlist([content, label], fields))\n",
    "        # 之前是一些预处理操作，此处调用super初始化父类，构造自定义的Dataset类\n",
    "        super().__init__(lists, fields)\n",
    "\n",
    "\n",
    "def read_data(path):\n",
    "    data_list = []\n",
    "    with open(path,encoding=\"utf-8\") as F:\n",
    "        for line in F:\n",
    "            line = line.split(\"\\t\")\n",
    "            text = line[0]\n",
    "            label = line[-1].strip().replace(' ','')\n",
    "            data_list.append((text,label))\n",
    "        return data_list\n",
    "           \n",
    "\n",
    "\n",
    "    \n",
    "def prepare_data_iter(config):\n",
    "    train_data_lists = read_data(config.training_path)\n",
    "    test_data_list = read_data(config.testing_path)\n",
    "    validation_data_list = read_data(config.validation_path)\n",
    "    \n",
    "    SRC = Field(tokenize=lambda x: x.split(), init_token=config.init_token, eos_token=config.eos_token,\n",
    "                fix_length=config.fix_length)\n",
    "    \n",
    "    SRC = Field(tokenize=lambda x: x.split(), \n",
    "                fix_length=config.fix_length)\n",
    "\n",
    "    LABEL = Field(tokenize=jieba.lcut, init_token=config.init_token, eos_token=config.eos_token,\n",
    "                  fix_length=config.fix_length)\n",
    "\n",
    "    train_dataset = MyDataset(train_data_lists, SRC, LABEL)\n",
    "    test_dataset = MyDataset(test_data_list, SRC, LABEL)\n",
    "    valid_dataset = MyDataset(validation_data_list,SRC,LABEL)\n",
    "\n",
    "#     SRC.build_vocab(train_dataset,min_freq = 2)  # '<unk>':0,'<sos>':1,'i':2\n",
    "#     LABEL.build_vocab(train_dataset,min_freq = 2)\n",
    "    SRC.build_vocab(train_dataset)  \n",
    "    LABEL.build_vocab(train_dataset)\n",
    "    \n",
    "    train_iter = data.BucketIterator(dataset=train_dataset, batch_size=config.batch_size,\n",
    "                                     shuffle=True, sort_key=lambda x: len(x.text),\n",
    "                                      sort_within_batch=False, repeat=False, device = config.device)\n",
    "\n",
    "    test_iter = data.BucketIterator(dataset=test_dataset, batch_size=config.batch_size,\n",
    "                                     shuffle=False, sort_key=lambda x: len(x.text),\n",
    "                                      sort_within_batch=False, repeat=False, device = config.device)\n",
    "    valid_iter = data.BucketIterator(dataset=valid_dataset, batch_size=config.batch_size,\n",
    "                                     shuffle=False, sort_key=lambda x: len(x.text),\n",
    "                                      sort_within_batch=False, repeat=False, device = config.device)\n",
    "    return train_iter, test_iter, valid_iter, SRC, LABEL"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 数字转文字"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.66886Z",
     "iopub.status.busy": "2021-10-23T02:27:54.668218Z",
     "iopub.status.idle": "2021-10-23T02:27:54.679616Z",
     "shell.execute_reply": "2021-10-23T02:27:54.678717Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.66881Z"
    }
   },
   "outputs": [],
   "source": [
    "def tokens2sentence(preds, LABEL):\n",
    "    sentences = []\n",
    "    for tokens in preds:\n",
    "        sentence = []\n",
    "        for token in tokens:\n",
    "            word = LABEL.vocab.itos[int(token.item())]\n",
    "            \n",
    "            if word == '<eos>':\n",
    "                break\n",
    "            sentence.append(word)\n",
    "        sentences.append(sentence)\n",
    "\n",
    "    return sentences"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 计算bleu score"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "https://blog.csdn.net/guolindonggld/article/details/56966200"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.681441Z",
     "iopub.status.busy": "2021-10-23T02:27:54.681017Z",
     "iopub.status.idle": "2021-10-23T02:27:54.690796Z",
     "shell.execute_reply": "2021-10-23T02:27:54.690047Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.681407Z"
    }
   },
   "outputs": [],
   "source": [
    "def computebleu(sentences, targets):\n",
    "    score = 0\n",
    "    count = len(sentences)\n",
    "    assert (len(sentences) == len(targets))\n",
    "\n",
    "    def cut_token(sentence):\n",
    "        tmp = []\n",
    "        for token in sentence:\n",
    "            if token == \"<eos>\":\n",
    "                break\n",
    "            if token == '<unk>' or token.isdigit() or len(bytes(token[0], encoding='utf-8')) == 1:\n",
    "                tmp.append(token)\n",
    "            else:\n",
    "                tmp += [word for word in token]\n",
    "        return tmp\n",
    "\n",
    "    for sentence, target in zip(sentences, targets):\n",
    "        sentence = cut_token(sentence)\n",
    "        target = cut_token(target)\n",
    "        score += sentence_bleu([target], sentence, weights=(1, 0, 0, 0))\n",
    "\n",
    "    return score, count"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 训练和测试"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.692566Z",
     "iopub.status.busy": "2021-10-23T02:27:54.692084Z",
     "iopub.status.idle": "2021-10-23T02:27:54.70134Z",
     "shell.execute_reply": "2021-10-23T02:27:54.700582Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.692533Z"
    }
   },
   "outputs": [],
   "source": [
    "def run_epoch(data_iter, model, loss_compute, pad):\n",
    "    \"\"\"\n",
    "    loss_compute：一个函数，包含全连接,计算损失，反向传播\n",
    "    \"\"\"\n",
    "    start = time.time()\n",
    "    total_tokens = 0\n",
    "    total_loss = 0\n",
    "    tokens = 0\n",
    "    for i, ((src,trg),_) in enumerate(data_iter):\n",
    "        src = src.permute(1,0)\n",
    "        trg = trg.permute(1,0)\n",
    "        batch = Batch(src, trg, pad)\n",
    "        \n",
    "        out = model.forward(batch.src, batch.trg, batch.src_mask, batch.trg_mask)\n",
    "        loss = loss_compute(out, batch.trg_y, batch.ntokens)\n",
    "        \n",
    "        total_loss += loss\n",
    "        total_tokens += batch.ntokens\n",
    "        tokens += batch.ntokens\n",
    "        if i % 50 == 1:\n",
    "            elapsed = time.time() - start\n",
    "            print(\"Epoch Step: %d Loss: %f Tokens per Sec: %f\" %\n",
    "                    (i, loss / batch.ntokens, tokens / elapsed))\n",
    "            start = time.time()\n",
    "            tokens = 0\n",
    "            \n",
    "    return total_loss / total_tokens"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.702912Z",
     "iopub.status.busy": "2021-10-23T02:27:54.702659Z",
     "iopub.status.idle": "2021-10-23T02:27:54.716126Z",
     "shell.execute_reply": "2021-10-23T02:27:54.715404Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.70288Z"
    }
   },
   "outputs": [],
   "source": [
    "def train_process(config, SRC, LABEL):\n",
    "\n",
    "    pad = SRC.vocab.stoi['<pad>']\n",
    "    src_vocab = len(SRC.vocab)\n",
    "    tgt_vocab = len(LABEL.vocab)\n",
    "\n",
    "    N = config.n_layers\n",
    "    d_model = config.d_model\n",
    "    d_ff = config.d_ff\n",
    "\n",
    "    h = config.h\n",
    "    dropout = config.dropout\n",
    "    device = config.device\n",
    "    model = make_model(src_vocab, tgt_vocab, N, d_model, d_ff, h, dropout).to(device)\n",
    "    \n",
    "    max_len = config.fix_length\n",
    "    start_symbol = LABEL.vocab.stoi['<bos>']\n",
    "    beam_size = config.beam_size\n",
    "    \n",
    "\n",
    "    # 损失函数\n",
    "    smoothing = config.smoothing\n",
    "    criterion = LabelSmoothing(tgt_vocab, pad, smoothing)\n",
    "\n",
    "    # 优化器\n",
    "    factor = config.factor\n",
    "    warmup = config.warmup\n",
    "    optimizer = torch.optim.Adam(model.parameters(),lr=0, betas = (0.9, 0.98), eps=1e-9)\n",
    "    model_opt = NoamOpt(d_model, factor, warmup, optimizer)\n",
    "\n",
    "    best_score = 0\n",
    "    # 封装计算损失函数的对象\n",
    "    loss_compute = SimpleLossCompute(model.generator, criterion, model_opt)\n",
    "    for i in range(config.epochs):\n",
    "        \n",
    "        run_epoch(train_iter, model, loss_compute, pad)\n",
    "        total_score, total_num, _ = test(model, valid_iter, pad, max_len, start_symbol,device, beam_size)\n",
    "        \n",
    "        print('epoch{}:total bleu score:{:.4f}, total scentence:{},average score{:.4f}'.format(\n",
    "            i,total_score,total_num,total_score/total_num))\n",
    "        if total_score/total_num > best_score:\n",
    "            best_score = total_score/total_num\n",
    "            # 保存模型\n",
    "            torch.save(model.state_dict(),'transformer.pth')\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.717749Z",
     "iopub.status.busy": "2021-10-23T02:27:54.717428Z",
     "iopub.status.idle": "2021-10-23T02:27:54.730944Z",
     "shell.execute_reply": "2021-10-23T02:27:54.730137Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.717714Z"
    }
   },
   "outputs": [],
   "source": [
    "def test(model, data_iter, pad, max_len, start_symbol, device, beam_size = 1):\n",
    "    model = model\n",
    "    total_score = 0\n",
    "    total_count = 0\n",
    "    results = []\n",
    "    print('testing...')\n",
    "    for (src, trg), _ in tqdm(data_iter):\n",
    "        \n",
    "        src = src.permute(1,0)\n",
    "        trg = trg.permute(1,0)\n",
    "        src_mask = (src != pad).unsqueeze(-2)\n",
    "        \n",
    "        if beam_size == 1:\n",
    "            ids = greedy_decode(model, src, src_mask, max_len, start_symbol)\n",
    "            sentences = tokens2sentence(ids, LABEL)\n",
    "            # label_text = tokens2sentence(ids, LABEL)\n",
    "        else:\n",
    "            ids = beam_search(model, src, src_mask, max_len, start_symbol, beam_size, device)\n",
    "            sentences = tokens2sentence(ids, LABEL)\n",
    "            \n",
    "        targets = tokens2sentence(trg, LABEL)\n",
    "        source = tokens2sentence(src, SRC)\n",
    "#         results.append((source,targets,sentences))\n",
    "        score, count = computebleu(sentences, targets)\n",
    "        for src, tgt, pred in zip(source, targets,sentences):\n",
    "            results.append((list(filter(lambda x: x!= '<pad>', src)), tgt, pred ))\n",
    "        total_score += score\n",
    "        total_count += count\n",
    "    return total_score, total_count, results\n",
    "                \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.73449Z",
     "iopub.status.busy": "2021-10-23T02:27:54.734092Z",
     "iopub.status.idle": "2021-10-23T02:27:54.74358Z",
     "shell.execute_reply": "2021-10-23T02:27:54.742764Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.734462Z"
    }
   },
   "outputs": [],
   "source": [
    "def load_model(config, LABEL, SRC):\n",
    "    pad = SRC.vocab.stoi['<pad>']\n",
    "    src_vocab = len(SRC.vocab)\n",
    "    tgt_vocab = len(LABEL.vocab)\n",
    "\n",
    "    N = config.n_layers\n",
    "    d_model = config.d_model\n",
    "    d_ff = config.d_ff\n",
    "\n",
    "    h = config.h\n",
    "    dropout = config.dropout\n",
    "    device = config.device\n",
    "    model = make_model(src_vocab, tgt_vocab, N, d_model, d_ff, h, dropout).to(device)\n",
    "    model.load_state_dict(torch.load(config.load_path))\n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## main"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:54.747189Z",
     "iopub.status.busy": "2021-10-23T02:27:54.746525Z",
     "iopub.status.idle": "2021-10-23T02:27:57.622989Z",
     "shell.execute_reply": "2021-10-23T02:27:57.622205Z",
     "shell.execute_reply.started": "2021-10-23T02:27:54.747163Z"
    }
   },
   "outputs": [],
   "source": [
    "config = Config()\n",
    "config.beam_size = 1\n",
    "train_iter, test_iter, valid_iter, SRC, LABEL = prepare_data_iter(config)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:27:57.624742Z",
     "iopub.status.busy": "2021-10-23T02:27:57.624324Z",
     "iopub.status.idle": "2021-10-23T02:50:02.120125Z",
     "shell.execute_reply": "2021-10-23T02:50:02.119417Z",
     "shell.execute_reply.started": "2021-10-23T02:27:57.624707Z"
    }
   },
   "outputs": [],
   "source": [
    "# 训练\n",
    "model = train_process(config, SRC, LABEL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T02:50:02.122286Z",
     "iopub.status.busy": "2021-10-23T02:50:02.121803Z",
     "iopub.status.idle": "2021-10-23T03:26:31.447529Z",
     "shell.execute_reply": "2021-10-23T03:26:31.446699Z",
     "shell.execute_reply.started": "2021-10-23T02:50:02.12225Z"
    }
   },
   "outputs": [],
   "source": [
    "config.load_path = '../input/epoch-20-smooth-02/transformer.pth'\n",
    "config.load_path = 'transformer.pth'\n",
    "# config = Config()\n",
    "# # 加载训练好的模型做预测\n",
    "model = load_model(config, LABEL, SRC)\n",
    "config.beam_size = 3\n",
    "    \n",
    "pad = SRC.vocab.stoi['<pad>']\n",
    "max_len = config.fix_length\n",
    "start_symbol = LABEL.vocab.stoi['<bos>']\n",
    "beam_size = config.beam_size\n",
    "device = config.device\n",
    "total_score, total_scentence, res = test(model, test_iter, pad, max_len, start_symbol, device, beam_size )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T03:26:31.449233Z",
     "iopub.status.busy": "2021-10-23T03:26:31.448968Z",
     "iopub.status.idle": "2021-10-23T03:26:31.455011Z",
     "shell.execute_reply": "2021-10-23T03:26:31.454336Z",
     "shell.execute_reply.started": "2021-10-23T03:26:31.4492Z"
    }
   },
   "outputs": [],
   "source": [
    "'''\n",
    "0.2\n",
    "(791.7836494832305, 2636)\n",
    "0.1\n",
    "(798.8936338983596, 2636)\n",
    "0\n",
    "(769.3034688196138, 2636)\n",
    "'''"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T03:26:31.457794Z",
     "iopub.status.busy": "2021-10-23T03:26:31.456032Z",
     "iopub.status.idle": "2021-10-23T03:26:31.471145Z",
     "shell.execute_reply": "2021-10-23T03:26:31.470248Z",
     "shell.execute_reply.started": "2021-10-23T03:26:31.457743Z"
    }
   },
   "outputs": [],
   "source": [
    "total_score, total_scentence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T03:26:31.473891Z",
     "iopub.status.busy": "2021-10-23T03:26:31.473635Z",
     "iopub.status.idle": "2021-10-23T03:26:31.479714Z",
     "shell.execute_reply": "2021-10-23T03:26:31.479017Z",
     "shell.execute_reply.started": "2021-10-23T03:26:31.473856Z"
    }
   },
   "outputs": [],
   "source": [
    "from IPython.display import FileLink\n",
    "FileLink(r'./transformer.pth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-23T03:26:31.48145Z",
     "iopub.status.busy": "2021-10-23T03:26:31.480849Z",
     "iopub.status.idle": "2021-10-23T03:26:33.517346Z",
     "shell.execute_reply": "2021-10-23T03:26:33.516819Z",
     "shell.execute_reply.started": "2021-10-23T03:26:31.481415Z"
    }
   },
   "outputs": [],
   "source": [
    "for source, target, pred in res:\n",
    "    print('-'*30)\n",
    "    print('原文：',' '.join(source))\n",
    "    print('预测：',''.join(pred))\n",
    "    print('目标：',''.join(target))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "warmup 展示\n",
    "```pyhton\n",
    "import torch\n",
    "import numpy as np\n",
    "import random\n",
    "from torch.utils.data import DataLoader, Dataset, TensorDataset\n",
    "from torch import optim\n",
    "from torch import nn\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "\n",
    "def setup_seed(seed):\n",
    "    torch.manual_seed(seed)\n",
    "    torch.cuda.manual_seed_all(seed)\n",
    "    np.random.seed(seed)\n",
    "    random.seed(seed)\n",
    "    torch.backends.cudnn.deterministic = True\n",
    "\n",
    "\n",
    "# 设置随机数种子\n",
    "setup_seed(20)\n",
    "\n",
    "\n",
    "class TensorDataset(Dataset):\n",
    "    \"\"\"Dataset wrapping data and target tensors.\n",
    "\n",
    "    Each sample will be retrieved by indexing both tensors along the first\n",
    "    dimension.\n",
    "\n",
    "    Arguments:\n",
    "        data_tensor (Tensor): contains sample data.\n",
    "        target_tensor (Tensor): contains sample targets (labels).\n",
    "    可以重写改成适合的方式\n",
    "    \"\"\"\n",
    "\n",
    "    def __init__(self, data_tensor, target_tensor):\n",
    "        assert data_tensor.size(0) == target_tensor.size(0)\n",
    "        self.data_tensor = data_tensor\n",
    "        self.target_tensor = target_tensor\n",
    "\n",
    "    def __getitem__(self, index):\n",
    "        return self.data_tensor[index], self.target_tensor[index]\n",
    "\n",
    "    def __len__(self):\n",
    "        return self.data_tensor.size(0)\n",
    "\n",
    "\n",
    "class NoamOpt:\n",
    "    \"Optim wrapper that implements rate.\"\n",
    "\n",
    "    def __init__(self, model_size, factor, warmup, optimizer):\n",
    "        self.optimizer = optimizer\n",
    "        self._step = 0\n",
    "        self.warmup = warmup\n",
    "        self.factor = factor\n",
    "        self.model_size = model_size\n",
    "        self._rate = 0\n",
    "\n",
    "    def step(self):\n",
    "        \"Update parameters and rate\"\n",
    "        self._step += 1\n",
    "        rate = self.rate()\n",
    "        for p in self.optimizer.param_groups:\n",
    "            p['lr'] = rate\n",
    "        self._rate = rate\n",
    "        self.optimizer.step()\n",
    "\n",
    "    def zero_grad(self):\n",
    "        self.optimizer.zero_grad()\n",
    "\n",
    "    def rate(self, step=None):\n",
    "        \"Implement `lrate` above\"\n",
    "        if step is None:\n",
    "            step = self._step\n",
    "        return self.factor * \\\n",
    "               (self.model_size ** (-0.5) *\n",
    "                min(step ** (-0.5), step * self.warmup ** (-1.5)))\n",
    "\n",
    "\n",
    "class Model(nn.Module):\n",
    "    def __init__(self, in_size, out_size):\n",
    "        super(Model, self).__init__()\n",
    "        self.fc = nn.Linear(in_size, out_size)\n",
    "\n",
    "    def forward(self, x):\n",
    "        out = self.fc(x)\n",
    "        return out\n",
    "\n",
    "\n",
    "def train(model, data_iter, optimizer, loss_function):\n",
    "    model.train()\n",
    "    for x, y in data_iter:\n",
    "        out = model(x)\n",
    "        optimizer.zero_grad()\n",
    "        loss = loss_function(out, y)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "\n",
    "    return model, optimizer\n",
    "\n",
    "\n",
    "def test(model, data_iter):\n",
    "    total_num = 0\n",
    "    acc_num = 0\n",
    "    model.eval()\n",
    "    for x, y in data_iter:\n",
    "        out = model(x)\n",
    "        _, pred = torch.max(out, dim=-1)\n",
    "        total_num += x.shape[0]\n",
    "        acc_num += sum(pred == y)\n",
    "    print('acc', acc_num / total_num)\n",
    "    return model, optimizer\n",
    "\n",
    "\n",
    "if __name__ == '__main__':\n",
    "\n",
    "    opts = [NoamOpt(10, 1, 20, None),\n",
    "            NoamOpt(10, 1, 40, None),\n",
    "            NoamOpt(20, 1, 20, None)]\n",
    "    plt.plot(np.arange(1, 100), [[opt.rate(i) for opt in opts] for i in range(1, 100)])\n",
    "    plt.legend([\"10:20\", \"10:40\", \"20:20\"])\n",
    "    None\n",
    "\n",
    "    plt.show()\n",
    "\n",
    "    # 构造训练集\n",
    "    d = 10  # 特征维度\n",
    "    n = 50  # 数据量\n",
    "\n",
    "    X = torch.randn(n, d)  # shape = 50,10\n",
    "    true_w = torch.randn(d, 1)\n",
    "    Y = X @ true_w + torch.randn(n, 1) * 0.1\n",
    "    Y[Y >= 0] = 1\n",
    "    Y[Y < 0] = 0\n",
    "\n",
    "    Y = Y.squeeze().long()\n",
    "    dataset = TensorDataset(X, Y)\n",
    "    loader = DataLoader(dataset=dataset, batch_size=8, shuffle=True, num_workers=2, )\n",
    "\n",
    "    model = Model(10, 2)\n",
    "    optimizer = optim.Adam(model.parameters(), lr=0, betas=(0.9, 0.98), eps=1e-9)\n",
    "    warm_up = NoamOpt(10, 2, 16, optimizer)\n",
    "    # warm_up = optimizer = optim.Adam(model.parameters(), lr=0.1, betas=(0.9, 0.98), eps=1e-9)\n",
    "    loss_function = nn.CrossEntropyLoss()\n",
    "\n",
    "    for i in range(50):\n",
    "        model, warm_up = train(model, loader, warm_up, loss_function)\n",
    "        test(model, loader)\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "标签平滑展示\n",
    "```python\n",
    "from torch import nn\n",
    "import torch\n",
    "from torch.autograd import Variable\n",
    "\n",
    "\n",
    "class LabelSmoothing(nn.Module):\n",
    "    \"Implement label smoothing.\"\n",
    "\n",
    "    def __init__(self, size, smoothing=0.0, padding_idx=None):\n",
    "        '''\n",
    "        :param size: 标签种类\n",
    "        :param padding_idx: 忽略某个标签\n",
    "        :param smoothing: 0-1   0 表示不进行标签平滑 1 表示完全平滑\n",
    "        '''\n",
    "        super(LabelSmoothing, self).__init__()\n",
    "        self.criterion = nn.KLDivLoss(size_average=False)\n",
    "        self.padding_idx = padding_idx\n",
    "        self.confidence = 1.0 - smoothing\n",
    "        self.smoothing = smoothing\n",
    "        self.size = size\n",
    "        self.true_dist = None\n",
    "\n",
    "    def forward(self, x, target):\n",
    "        assert x.size(1) == self.size\n",
    "        true_dist = x.data.clone()\n",
    "        # 把平滑的可能性平均分给除了pad 和 真实的标签\n",
    "        if self.padding_idx is not None:\n",
    "            true_dist.fill_(self.smoothing / (self.size - 2))\n",
    "        else:\n",
    "            true_dist.fill_(self.smoothing / (self.size - 1))\n",
    "        true_dist.scatter_(1, target.data.unsqueeze(1), self.confidence)\n",
    "        if self.padding_idx is not None:\n",
    "            true_dist[:, self.padding_idx] = 0\n",
    "            mask = torch.nonzero(target.data == self.padding_idx)\n",
    "            if mask.dim() > 0:\n",
    "                '''\n",
    "                把真实标签为padding_idx的标签的所有都调成0,这样pad就不会影响损失\n",
    "                最开始是\n",
    "                [2, 1, 0]变为\n",
    "                本来是tensor([[0.0000, 0.1333, 0.6000, 0.1333, 0.1333],\n",
    "                [0.0000, 0.6000, 0.1333, 0.1333, 0.1333],\n",
    "                [0.0000, 0.1333, 0.1333, 0.1333, 0.1333]])\n",
    "                ---->\n",
    "                tensor([[0.0000, 0.1333, 0.6000, 0.1333, 0.1333],\n",
    "                [0.0000, 0.6000, 0.1333, 0.1333, 0.1333],\n",
    "                [0.0000, 0.0000, 0.0000, 0.0000, 0.0000]])\n",
    "                '''\n",
    "                true_dist.index_fill_(0, mask.squeeze(), 0.0)\n",
    "        self.true_dist = true_dist\n",
    "        return self.criterion(x, Variable(true_dist, requires_grad=False))\n",
    "\n",
    "\n",
    "predict = torch.FloatTensor([[0, 0.2, 0.7, 0.1, 0],\n",
    "                             [0, 0.2, 0.7, 0.1, 0],\n",
    "                             [0, 0.2, 0.7, 0.1, 0]])\n",
    "\n",
    "label = Variable(torch.LongTensor([2, 1, 0]))\n",
    "\n",
    "crit = LabelSmoothing(5, 0.4, 0)\n",
    "\n",
    "print(crit(predict, label))\n",
    "# 验证第三条数据对损失没有任何贡献\n",
    "predict = torch.FloatTensor([[0, 0.2, 0.7, 0.1, 0],\n",
    "                             [0, 0.2, 0.7, 0.1, 0],\n",
    "                             [0, 0, 0, 0, 0]])\n",
    "\n",
    "print(crit(predict, label))\n",
    "\n",
    "# 让pad 影响损失\n",
    "crit = LabelSmoothing(5, 0.4)\n",
    "predict = torch.FloatTensor([[0, 0.2, 0.7, 0.1, 0],\n",
    "                             [0, 0.2, 0.7, 0.1, 0],\n",
    "                             [0, 0.2, 0.7, 0.1, 0]])\n",
    "print(crit(predict, label))\n",
    "\n",
    "torch.LongTensor\n",
    "\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}