深度学习——tensorflow2.0搭建文本分类模型
import matplotlib as mpl
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import sklearn
import pandas as pd
import os
import sys
import time
import tensorflow as tffrom tensorflow import keras
print(tf.__version__)
2.0.0-alpha0
imdb = keras.datasets.imdb
vocab_size = 10000
index_from = 3
# num_words：词表的最大值；index_from：从这个index开始索引实际单词，默认为3
(train_data, train_labels), (test_data, test_labels) = imdb.load_data(num_words = vocab_size, index_from = index_from)word_index = imdb.get_word_index()
print(len(word_index))
88584
word_index = {k:(v+3) for k, v in word_index.items()}
word_index[''] = 0
word_index[''] = 1
word_index[''] = 2
word_index[''] = 3
reverse_word_index = dict([(value, key) for key, value in word_index.items()])
def decode_review(text_ids):
return ' '.join([reverse_word_index.get(word_id, "") for word_id in text_ids])
max_length = 500
# keras.preprocessing.sequence.pad_sequences:由于输入是变长的，因此需要使用padding进行对齐
train_data = keras.preprocessing.sequence.pad_sequences(
train_data, # list of list
value = word_index[''], # padding填充的数值
padding = 'post', # post, pre:post表示在句子的后面进行padding，pre表示在句子的前面进行padding
maxlen = max_length)test_data = keras.preprocessing.sequence.pad_sequences(
test_data, # list of list
value = word_index[''],
padding = 'post', # post, pre
maxlen = max_length)
embedding_dim = 16
batch_size = 512bi_rnn_model = keras.models.Sequential([
keras.layers.Embedding(vocab_size, embedding_dim, input_length = max_length),
keras.layers.Bidirectional(keras.layers.LSTM(units = 32, return_sequences = False)),
keras.layers.Dense(32, activation = 'relu'),
keras.layers.Dense(1, activation='sigmoid'),
])bi_rnn_model.summary()
# binary_crossentropy：2分类的交叉熵损失函数
bi_rnn_model.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])
Model: "sequential_2" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= embedding_2 (Embedding) (None, 500, 16) 160000 _________________________________________________________________ bidirectional_2 (Bidirection (None, 64) 12544 _________________________________________________________________ dense_4 (Dense) (None, 32) 2080 _________________________________________________________________ dense_5 (Dense) (None, 1) 33 ================================================================= Total params: 174,657 Trainable params: 174,657 Non-trainable params: 0 _________________________________________________________________
history = bi_rnn_model.fit(train_data, train_labels, epochs = 30, batch_size = batch_size, validation_split = 0.2)
Train on 20000 samples, validate on 5000 samples Epoch 1/30 20000/20000 [==============================] - 6s 307us/sample - loss: 0.6858 - accuracy: 0.5631 - val_loss: 0.6281 - val_accuracy: 0.6702 Epoch 2/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.4969 - accuracy: 0.7776 - val_loss: 0.4603 - val_accuracy: 0.7960 Epoch 3/30 20000/20000 [==============================] - 6s 281us/sample - loss: 0.3552 - accuracy: 0.8547 - val_loss: 0.3183 - val_accuracy: 0.8732 Epoch 4/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.2202 - accuracy: 0.9201 - val_loss: 0.2940 - val_accuracy: 0.8834 Epoch 5/30 20000/20000 [==============================] - 6s 281us/sample - loss: 0.1672 - accuracy: 0.9452 - val_loss: 0.2979 - val_accuracy: 0.8882 Epoch 6/30 20000/20000 [==============================] - 6s 281us/sample - loss: 0.1301 - accuracy: 0.9603 - val_loss: 0.3593 - val_accuracy: 0.8812 Epoch 7/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.1052 - accuracy: 0.9693 - val_loss: 0.3746 - val_accuracy: 0.8770 Epoch 8/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0886 - accuracy: 0.9754 - val_loss: 0.3826 - val_accuracy: 0.8736 Epoch 9/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0824 - accuracy: 0.9772 - val_loss: 0.3935 - val_accuracy: 0.8826 Epoch 10/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0675 - accuracy: 0.9822 - val_loss: 0.4086 - val_accuracy: 0.8726 Epoch 11/30 20000/20000 [==============================] - 6s 281us/sample - loss: 0.0873 - accuracy: 0.9715 - val_loss: 0.5023 - val_accuracy: 0.8718 Epoch 12/30 20000/20000 [==============================] - 6s 281us/sample - loss: 0.0613 - accuracy: 0.9829 - val_loss: 0.4441 - val_accuracy: 0.8796 Epoch 13/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0534 - accuracy: 0.9857 - val_loss: 0.4314 - val_accuracy: 0.8684 Epoch 14/30 20000/20000 [==============================] - 6s 281us/sample - loss: 0.0512 - accuracy: 0.9864 - val_loss: 0.5086 - val_accuracy: 0.8728 Epoch 15/30 20000/20000 [==============================] - 6s 281us/sample - loss: 0.0375 - accuracy: 0.9920 - val_loss: 0.5097 - val_accuracy: 0.8728 Epoch 16/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0384 - accuracy: 0.9915 - val_loss: 0.5945 - val_accuracy: 0.8666 Epoch 17/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0331 - accuracy: 0.9929 - val_loss: 0.5916 - val_accuracy: 0.8714 Epoch 18/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0263 - accuracy: 0.9955 - val_loss: 0.6204 - val_accuracy: 0.8724 Epoch 19/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0234 - accuracy: 0.9961 - val_loss: 0.5929 - val_accuracy: 0.8722 Epoch 20/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0193 - accuracy: 0.9972 - val_loss: 0.6174 - val_accuracy: 0.8706 Epoch 21/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0173 - accuracy: 0.9976 - val_loss: 0.6405 - val_accuracy: 0.8696 Epoch 22/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0157 - accuracy: 0.9977 - val_loss: 0.6853 - val_accuracy: 0.8682 Epoch 23/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0161 - accuracy: 0.9973 - val_loss: 0.7232 - val_accuracy: 0.8564 Epoch 24/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0751 - accuracy: 0.9758 - val_loss: 0.4803 - val_accuracy: 0.8574 Epoch 25/30 20000/20000 [==============================] - 6s 279us/sample - loss: 0.0354 - accuracy: 0.9898 - val_loss: 0.6191 - val_accuracy: 0.8686 Epoch 26/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0196 - accuracy: 0.9955 - val_loss: 0.6793 - val_accuracy: 0.8628 Epoch 27/30 20000/20000 [==============================] - 6s 283us/sample - loss: 0.0150 - accuracy: 0.9974 - val_loss: 0.7201 - val_accuracy: 0.8654 Epoch 28/30 20000/20000 [==============================] - 6s 287us/sample - loss: 0.0137 - accuracy: 0.9974 - val_loss: 0.7434 - val_accuracy: 0.8648 Epoch 29/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.0232 - accuracy: 0.9936 - val_loss: 0.9976 - val_accuracy: 0.8578 Epoch 30/30 20000/20000 [==============================] - 6s 280us/sample - loss: 0.1400 - accuracy: 0.9535 - val_loss: 0.4958 - val_accuracy: 0.8592
bi_rnn_model.evaluate(test_data, test_labels, batch_size = batch_size)
25000/25000 [==============================] - 2s 99us/sample - loss: 0.5320 - accuracy: 0.8460
Out[ ]:
[0.5319758009338379, 0.846]