TextCNN的完整步骤（不到60行代码）

前言：

TextCNN对于字符串长度不是很长（20个字符以下）的文本分类效果较好，比如要通过新闻标题对新闻种类进行分类，就可以使用TextCNN，主要的步骤如下：
1、导入数据 read_csv
2、jieba分词（中文）
3、建立词库 Tokenizer.fit_on_texts
4、把词转换成数字编号 Tokenizer.texts_to_sequences
5、规整单个sentence的长度 pad_sequences
6、构建textcnn主函数并训练
7、predict出最后结果

注：我也没有数据集，可以自己爬豆瓣玩玩

import tensorflow as tf
import pandas as pd
import jieba
from tensorflow.keras import layers
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences
from sklearn.model_selection import train_test_split

#数据导入和预处理
data = pd.read_csv('data.csv')
data['content'] = data['content'].apply(lambda x: list(jieba.cut(x)))
tk = Tokenizer()
tk.fit_on_texts(data['content'])
feature_nums = len(tk.word_index) + 1
train_x, test_x, train_y, test_y = train_test_split(data['content'], data['label'], random_state=10, test_size=0.2)
train_x = tk.texts_to_sequences(train_x)
test_x = tk.texts_to_sequences(test_x)

out_dimension = 300
sentence_length = 100

train_x = pad_sequences(train_x, maxlen=sentence_length)
test_x = pad_sequences(test_x, maxlen=sentence_length)

#编制模型
def convolutions():
    '''textcnn的多个卷积'''
    inpt = layers.Input(shape=(sentence_length, out_dimension, 1))
    out_list = []
    for size in [3,4,5]:
        conv = layers.Conv2D(filters=64, kernel_size=(size, out_dimension), activation='relu', strides=1)(inpt)
        conv = layers.MaxPool2D(pool_size=(sentence_length-size+1,1))(conv)
        out_list.append(conv)
    outt = layers.concatenate(out_list)
    model = tf.keras.Model(inputs=inpt, outputs=outt)
    return model

def textcnn(feature_nums, out_dimension, sentence_length):
    '''textcnn的主函数'''
    model = tf.keras.Sequential([
        layers.Embedding(input_dim=feature_nums, output_dim=out_dimension, input_length=sentence_length),
        layers.Reshape((sentence_length, out_dimension, 1)),
        convolutions(),
        layers.Flatten(),
        layers.Dense(100, activation='relu'),
        layers.Dropout(0.1),
        layers.Dense(3, activation='softmax')
    ])
    model.compile(optimizer=tf.keras.optimizers.Adam(), loss=tf.keras.losses.SparseCategoricalCrossentropy(),metrics=['accuracy'])
    return model

#模型训练和预测
model = textcnn(feature_nums, out_dimension, sentence_length)
model.fit(train_x, train_y, batch_size=64, epochs=10, validation_split=0.2)
y_pred = model.predict_classes(test_x)