{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "10291b0e", "metadata": {}, "outputs": [], "source": [ "import matplotlib as mpl\n", "import matplotlib.pyplot as plt\n", "import numpy as np\n", "import sklearn\n", "import pandas as pd\n", "import os\n", "import sys\n", "import time\n", "import tensorflow as tf\n", "from tensorflow import keras\n", "from tensorflow.keras.models import Model" ] }, { "cell_type": "code", "execution_count": 2, "id": "2e43f504", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "2.10.0\n", "sys.version_info(major=3, minor=9, micro=15, releaselevel='final', serial=0)\n", "matplotlib 3.6.2\n", "numpy 1.23.4\n", "pandas 1.5.1\n", "sklearn 1.1.3\n", "tensorflow 2.10.0\n", "keras.api._v2.keras 2.10.0\n" ] } ], "source": [ "# 查看库信息\n", "print(tf.__version__)\n", "print(sys.version_info)\n", "for module in mpl, np, pd, sklearn, tf, keras:\n", " print(module.__name__, module.__version__)" ] }, { "cell_type": "code", "execution_count": 3, "id": "8a0e2b33", "metadata": {}, "outputs": [], "source": [ "# 一,处理数据\n", "# 1,加载训练数据\n", "# https://storage.googleapis.com/download.tensorflow.org/data/shakespeare.txt\n", "input_filepath = \"./shakespeare.txt\"\n", "text = open(input_filepath, 'r').read()\n", "# 简单输出查看数据\n", "# print(len(text))\n", "# print(text[0:100])" ] }, { "cell_type": "code", "execution_count": 4, "id": "bdff13a8", "metadata": {}, "outputs": [], "source": [ "# 2,生成词库\n", "vocab = sorted(set(text))\n", "# 查看生成的词库\n", "# print(len(vocab))\n", "# print(vocab)" ] }, { "cell_type": "code", "execution_count": 5, "id": "17aee416", "metadata": {}, "outputs": [], "source": [ "# 3,生成由 字符->数字 的字典\n", "char2idx = {char:idx for idx, char in enumerate(vocab)}\n", "# 查看字典\n", "# print(char2idx)" ] }, { "cell_type": "code", "execution_count": 6, "id": "ffdf1013", "metadata": {}, "outputs": [], "source": [ "# 4,将词库转换为numpy中的array\n", "idx2char = np.array(vocab)\n", "# print(idx2char)" ] }, { "cell_type": "code", "execution_count": 7, "id": "6ceb6262", "metadata": {}, "outputs": [], "source": [ "# 5,将文本转化为数字\n", "text_as_int = np.array([char2idx[c] for c in text])\n", "# print(text_as_int[0:10])\n", "# print(text[0:10])" ] }, { "cell_type": "code", "execution_count": 8, "id": "8d8dbedd", "metadata": { "scrolled": true }, "outputs": [], "source": [ "# 6,将数据加载到dataset中,并处理数据\n", "\n", "# 将array转换为tensor格式的数据\n", "char_dataset = tf.data.Dataset.from_tensor_slices(text_as_int)\n", "\n", "# 使用tensor的方法将数据集每个100个字符进行batch分序列\n", "seq_length = 100\n", "seq_dataset = char_dataset.batch(seq_length + 1, drop_remainder = True)\n", "# 选择数据查看\n", "# for ch_id in char_dataset.take(2):\n", "# print(ch_id, idx2char[ch_id.numpy()])\n", "\n", "# for seq_id in seq_dataset.take(2):\n", "# print(seq_id)\n", "# print(repr(''.join(idx2char[seq_id.numpy()])))" ] }, { "cell_type": "code", "execution_count": 9, "id": "2c02cfa1", "metadata": {}, "outputs": [], "source": [ "# 7,将数据分割成输入和输出两个部分\n", "def split_input_target(id_text):\n", " # 将序列作如下变化 abcde -> abcd, bcde\n", " return id_text[0:-1], id_text[1:]\n", "seq_dataset = seq_dataset.map(split_input_target)\n", "# 简单查看结果\n", "# for item_input, item_output in seq_dataset.take(2):\n", "# print(item_input.numpy())\n", "# print(item_output.numpy())" ] }, { "cell_type": "code", "execution_count": 10, "id": "ebb8f89c", "metadata": {}, "outputs": [], "source": [ "# 8,打乱数据,batch分组,batch_size=64\n", "batch_size = 64\n", "buffer_size = 10000\n", "\n", "seq_dataset = seq_dataset.shuffle(buffer_size).batch(batch_size, drop_remainder=True)" ] }, { "cell_type": "code", "execution_count": 11, "id": "5b91648e", "metadata": {}, "outputs": [], "source": [ "# 二,构建模型\n", "# 1,定义模型常量\n", "vocab_size = len(vocab)\n", "embedding_dim = 256\n", "rnn_units = 1024" ] }, { "cell_type": "code", "execution_count": 12, "id": "af8c56b7", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "(64, None)\n", "(64, None, 256)\n", "(64, None, 1024)\n", "(64, None, 65)\n", "Model: \"model\"\n", "_________________________________________________________________\n", " Layer (type) Output Shape Param # \n", "=================================================================\n", " input_1 (InputLayer) [(64, None)] 0 \n", " \n", " embedding (Embedding) (64, None, 256) 16640 \n", " \n", " lstm (LSTM) (64, None, 1024) 5246976 \n", " \n", " dense (Dense) (64, None, 65) 66625 \n", " \n", "=================================================================\n", "Total params: 5,330,241\n", "Trainable params: 5,330,241\n", "Non-trainable params: 0\n", "_________________________________________________________________\n" ] } ], "source": [ "# 2,定义model模型\n", "inputs = keras.Input(batch_input_shape=(batch_size,None))\n", "print(inputs.shape)\n", "outputs = keras.layers.Embedding(vocab_size, embedding_dim)(inputs)\n", "print(outputs.shape)\n", "outputs = keras.layers.LSTM(units = rnn_units,stateful = True,recurrent_initializer='glorot_uniform',return_sequences = True)(outputs)\n", "print(outputs.shape)\n", "outputs = keras.layers.Dense(vocab_size)(outputs)\n", "print(outputs.shape)\n", "\n", "model = Model(inputs, outputs)\n", "model.summary()" ] }, { "cell_type": "code", "execution_count": 13, "id": "0392a946", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "(64, 100, 65)\n", "tf.Tensor(\n", "[[41]\n", " [46]\n", " [27]\n", " [41]\n", " [12]\n", " [26]\n", " [48]\n", " [ 0]\n", " [12]\n", " [33]\n", " [29]\n", " [50]\n", " [32]\n", " [18]\n", " [ 7]\n", " [28]\n", " [64]\n", " [25]\n", " [64]\n", " [ 8]\n", " [40]\n", " [21]\n", " [ 7]\n", " [34]\n", " [58]\n", " [37]\n", " [23]\n", " [11]\n", " [56]\n", " [ 3]\n", " [41]\n", " [42]\n", " [33]\n", " [ 9]\n", " [ 0]\n", " [41]\n", " [ 4]\n", " [62]\n", " [38]\n", " [58]\n", " [46]\n", " [64]\n", " [36]\n", " [15]\n", " [33]\n", " [14]\n", " [52]\n", " [53]\n", " [26]\n", " [12]\n", " [ 0]\n", " [28]\n", " [34]\n", " [39]\n", " [50]\n", " [51]\n", " [62]\n", " [14]\n", " [19]\n", " [16]\n", " [29]\n", " [ 7]\n", " [41]\n", " [10]\n", " [31]\n", " [52]\n", " [48]\n", " [21]\n", " [61]\n", " [23]\n", " [28]\n", " [26]\n", " [38]\n", " [ 6]\n", " [64]\n", " [62]\n", " [16]\n", " [10]\n", " [36]\n", " [27]\n", " [27]\n", " [64]\n", " [15]\n", " [64]\n", " [58]\n", " [35]\n", " [64]\n", " [ 9]\n", " [59]\n", " [25]\n", " [ 5]\n", " [11]\n", " [24]\n", " [42]\n", " [31]\n", " [50]\n", " [13]\n", " [30]\n", " [31]\n", " [51]], shape=(100, 1), dtype=int64)\n", "tf.Tensor(\n", "[41 46 27 41 12 26 48 0 12 33 29 50 32 18 7 28 64 25 64 8 40 21 7 34\n", " 58 37 23 11 56 3 41 42 33 9 0 41 4 62 38 58 46 64 36 15 33 14 52 53\n", " 26 12 0 28 34 39 50 51 62 14 19 16 29 7 41 10 31 52 48 21 61 23 28 26\n", " 38 6 64 62 16 10 36 27 27 64 15 64 58 35 64 9 59 25 5 11 24 42 31 50\n", " 13 30 31 51], shape=(100,), dtype=int64)\n", "Input: \"?\\nWhere are your mess of sons to back you now?\\nThe wanton Edward, and the lusty George?\\nAnd where's \"\n", "\n", "Output: \"\\nWhere are your mess of sons to back you now?\\nThe wanton Edward, and the lusty George?\\nAnd where's t\"\n", "\n", "Predictions: \"chOc?Nj\\n?UQlTF-PzMz.bI-VtYK;r$cdU3\\nc&xZthzXCUBnoN?\\nPValmxBGDQ-c:SnjIwKPNZ,zxD:XOOzCztWz3uM';LdSlARSm\"\n" ] } ], "source": [ "# 3,单个例子测试模型\n", "for input_example_batch, target_example_batch in seq_dataset.take(1):\n", " example_batch_predictions = model(input_example_batch)\n", " print(example_batch_predictions.shape)\n", "\n", "# random sampling.\n", "# greedy, random.\n", "# 测试单个例子的结果\n", "sample_indices = tf.random.categorical(logits = example_batch_predictions[0], num_samples = 1)\n", "print(sample_indices)\n", "# (100, 65) -> (100, 1)\n", "sample_indices = tf.squeeze(sample_indices, axis = -1)\n", "print(sample_indices)\n", "\n", "\n", "# 打印输入,目标,预测的结果\n", "print(\"Input: \", repr(\"\".join(idx2char[input_example_batch[0]])))\n", "print()\n", "print(\"Output: \", repr(\"\".join(idx2char[target_example_batch[0]])))\n", "print()\n", "print(\"Predictions: \", repr(\"\".join(idx2char[sample_indices])))" ] }, { "cell_type": "code", "execution_count": 14, "id": "2f4fbdfa", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "(64, 100)\n", "4.1738906\n" ] } ], "source": [ "# 三,定义损失函数和优化器\n", "def loss(labels, logits):\n", " return keras.losses.sparse_categorical_crossentropy(labels, logits, from_logits=True)\n", "\n", "# 定义优化器和自定义损失函数\n", "model.compile(optimizer = 'adam', loss = loss)\n", "\n", "# 测试计算单例的损失数\n", "example_loss = loss(target_example_batch, example_batch_predictions)\n", "print(example_loss.shape)\n", "print(example_loss.numpy().mean())" ] }, { "cell_type": "code", "execution_count": 15, "id": "485a24b4", "metadata": {}, "outputs": [], "source": [ "# 四,callback模块-checkpoints\n", "output_dir = \"./text_generation_lstm3_checkpoints\"\n", "if not os.path.exists(output_dir):\n", " os.mkdir(output_dir)\n", "checkpoint_prefix = os.path.join(output_dir, 'ckpt_{epoch}')\n", "checkpoint_callback = keras.callbacks.ModelCheckpoint(filepath = checkpoint_prefix,save_weights_only = True)" ] }, { "cell_type": "code", "execution_count": 18, "id": "ee0599cf", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Epoch 1/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.4082\n", "Epoch 2/100\n", "172/172 [==============================] - 10s 52ms/step - loss: 0.4061\n", "Epoch 3/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.4057\n", "Epoch 4/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.4055\n", "Epoch 5/100\n", "172/172 [==============================] - 9s 45ms/step - loss: 0.4052\n", "Epoch 6/100\n", "172/172 [==============================] - 9s 45ms/step - loss: 0.4036\n", "Epoch 7/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.4031\n", "Epoch 8/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.4041\n", "Epoch 9/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.4013\n", "Epoch 10/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.4015\n", "Epoch 11/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.4002\n", "Epoch 12/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3983\n", "Epoch 13/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3962\n", "Epoch 14/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3954\n", "Epoch 15/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3928\n", "Epoch 16/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3951\n", "Epoch 17/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3940\n", "Epoch 18/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3936\n", "Epoch 19/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3936\n", "Epoch 20/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3981\n", "Epoch 21/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3970\n", "Epoch 22/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3986\n", "Epoch 23/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3941\n", "Epoch 24/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3935\n", "Epoch 25/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3920\n", "Epoch 26/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3928\n", "Epoch 27/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3936\n", "Epoch 28/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3925\n", "Epoch 29/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3918\n", "Epoch 30/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3915\n", "Epoch 31/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3950\n", "Epoch 32/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3936\n", "Epoch 33/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3920\n", "Epoch 34/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3906\n", "Epoch 35/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3919\n", "Epoch 36/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3895\n", "Epoch 37/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3908\n", "Epoch 38/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3891\n", "Epoch 39/100\n", "172/172 [==============================] - 9s 46ms/step - loss: 0.3894\n", "Epoch 40/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3929\n", "Epoch 41/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3906\n", "Epoch 42/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3908\n", "Epoch 43/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3889\n", "Epoch 44/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3892\n", "Epoch 45/100\n", "172/172 [==============================] - 9s 47ms/step - loss: 0.3904\n", "Epoch 46/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3879\n", "Epoch 47/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3914\n", "Epoch 48/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3931\n", "Epoch 49/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3904\n", "Epoch 50/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3947\n", "Epoch 51/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3963\n", "Epoch 52/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3906\n", "Epoch 53/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3919\n", "Epoch 54/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3913\n", "Epoch 55/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.3887\n", "Epoch 56/100\n", "172/172 [==============================] - 9s 48ms/step - loss: 0.3910\n", "Epoch 57/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3914\n", "Epoch 58/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3924\n", "Epoch 59/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3957\n", "Epoch 60/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3945\n", "Epoch 61/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3931\n", "Epoch 62/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3941\n", "Epoch 63/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3962\n", "Epoch 64/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.3959\n", "Epoch 65/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.3964\n", "Epoch 66/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3965\n", "Epoch 67/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3946\n", "Epoch 68/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3970\n", "Epoch 69/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3971\n", "Epoch 70/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3963\n", "Epoch 71/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.3976\n", "Epoch 72/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.3978\n", "Epoch 73/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.3979\n", "Epoch 74/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4021\n", "Epoch 75/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4043\n", "Epoch 76/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4022\n", "Epoch 77/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4053\n", "Epoch 78/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4050\n", "Epoch 79/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4036\n", "Epoch 80/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4015\n", "Epoch 81/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4025\n", "Epoch 82/100\n", "172/172 [==============================] - 9s 49ms/step - loss: 0.4047\n", "Epoch 83/100\n", "172/172 [==============================] - 9s 51ms/step - loss: 0.4042\n", "Epoch 84/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4042\n", "Epoch 85/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4061\n", "Epoch 86/100\n", "172/172 [==============================] - 9s 51ms/step - loss: 0.4065\n", "Epoch 87/100\n", "172/172 [==============================] - 9s 51ms/step - loss: 0.4089\n", "Epoch 88/100\n", "172/172 [==============================] - 9s 51ms/step - loss: 0.4119\n", "Epoch 89/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4150\n", "Epoch 90/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4204\n", "Epoch 91/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4168\n", "Epoch 92/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4172\n", "Epoch 93/100\n", "172/172 [==============================] - 9s 51ms/step - loss: 0.4168\n", "Epoch 94/100\n", "172/172 [==============================] - 10s 51ms/step - loss: 0.4158\n", "Epoch 95/100\n", "172/172 [==============================] - 9s 51ms/step - loss: 0.4173\n", "Epoch 96/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4130\n", "Epoch 97/100\n", "172/172 [==============================] - 10s 52ms/step - loss: 0.4120\n", "Epoch 98/100\n", "172/172 [==============================] - 10s 52ms/step - loss: 0.4147\n", "Epoch 99/100\n", "172/172 [==============================] - 9s 51ms/step - loss: 0.4198\n", "Epoch 100/100\n", "172/172 [==============================] - 9s 50ms/step - loss: 0.4270\n" ] } ], "source": [ "# 五,训练模型\n", "epochs = 100\n", "history = model.fit(seq_dataset, epochs = epochs,callbacks = [checkpoint_callback])\n", "\n", "# 会自动找到最近保存的变量文件\n", "new_checkpoint = tf.train.latest_checkpoint(output_dir)" ] }, { "cell_type": "code", "execution_count": 16, "id": "805747b3", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "(1, None)\n", "(1, None, 256)\n", "(1, None, 1024)\n", "(1, None, 65)\n", "Model: \"model_1\"\n", "_________________________________________________________________\n", " Layer (type) Output Shape Param # \n", "=================================================================\n", " input_2 (InputLayer) [(1, None)] 0 \n", " \n", " embedding_1 (Embedding) (1, None, 256) 16640 \n", " \n", " lstm_1 (LSTM) (1, None, 1024) 5246976 \n", " \n", " dense_1 (Dense) (1, None, 65) 66625 \n", " \n", "=================================================================\n", "Total params: 5,330,241\n", "Trainable params: 5,330,241\n", "Non-trainable params: 0\n", "_________________________________________________________________\n" ] } ], "source": [ "# 六,定义预测模型\n", "# 1,使用checkpoint加载模型\n", "inputs = keras.Input(batch_input_shape=(1,None))\n", "print(inputs.shape)\n", "outputs = keras.layers.Embedding(vocab_size, embedding_dim)(inputs)\n", "print(outputs.shape)\n", "outputs = keras.layers.LSTM(units = rnn_units,stateful = True,recurrent_initializer='glorot_uniform',\n", " return_sequences = True)(outputs)\n", "print(outputs.shape)\n", "outputs = keras.layers.Dense(vocab_size)(outputs)\n", "print(outputs.shape)\n", "\n", "model2 = Model(inputs, outputs)\n", "model2.load_weights(tf.train.latest_checkpoint(output_dir))\n", "model2.summary()" ] }, { "cell_type": "code", "execution_count": 17, "id": "b6e201d7", "metadata": {}, "outputs": [], "source": [ "# 2.使用h5文件加载模型\n", "from keras.models import load_model\n", "model2 = load_model('model.h5', compile=False)" ] }, { "cell_type": "code", "execution_count": 56, "id": "d99e6d71", "metadata": { "scrolled": true }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "All kiss m\n" ] } ], "source": [ "# 七,预测模型做预测\n", "def generate_text(model, start_string, num_generate = 1000):\n", " input_eval = [char2idx[ch] for ch in start_string]\n", " input_eval = tf.expand_dims(input_eval, 0)\n", " \n", " text_generated = []\n", " model.reset_states()\n", " \n", " for _ in range(num_generate):\n", " # 输入进入模型，得到预测\n", " predictions = model(input_eval)\n", " # 去掉一维的结构\n", " predictions = tf.squeeze(predictions, 0)\n", " # 根据此位置的词库中不同字符的可能性进行采样，获得预测出的字符\n", " predicted_id = tf.random.categorical(predictions, num_samples = 1)[-1, 0].numpy()\n", " # 将预测出的字符最后一位添加到已经生成的字符串末尾\n", " text_generated.append(idx2char[predicted_id])\n", " # 将预测出的字符最后一位作为输入再次输入到模型中预测\n", " input_eval = tf.expand_dims([predicted_id], 0)\n", " return start_string + ''.join(text_generated)\n", "\n", "new_text = generate_text(model2, \"Al\",8)\n", "print(new_text)" ] }, { "cell_type": "code", "execution_count": 17, "id": "e6a63d0f", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "WARNING:tensorflow:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.\n" ] } ], "source": [ "# 将模型保存为文件，节省空间\n", "# model2.save(\"model.h5\")" ] }, { "cell_type": "code", "execution_count": null, "id": "d617fa97", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "TensorGPU", "language": "python", "name": "tensorgpu" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.9.15" } }, "nbformat": 4, "nbformat_minor": 5 }