Python使用Keras库中的LSTM模型生成新文本内容教程

# 下载小说文本库
import urllib.request
url = 'https://cloud.tencent.com/developer/attachments/1632706'
response = urllib.request.urlopen(url)
text = response.read().decode('utf-8')
print(text)

# 对文本进行处理
import string
from collections import Counter
import re
# 清理文本
text = text.lower()
text = re.sub(r'\W+', ' ', text)
text = text.split()

# 构建LSTM模型
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
from keras.layers import LSTM
from keras.callbacks import ModelCheckpoint
from keras.utils import np_utils
# 创建序列，将文本进行X, y的拆分
sequence_length = 100
sequences = []
for i in range(sequence_length, len(text)):
    seq = text[i-sequence_length:i+1]
    sequences.append(' '.join(seq))
# 构建字典
words = ' '.join(sequences).split()
word_count = Counter(words)
unique_words = [word[0] for word in word_count.most_common(2000)]
word_to_index = {}
index_to_word = {}
for i, word in enumerate(unique_words):
    word_to_index[word] = i
    index_to_word[i] = word
# 构建x和y
X = []
y = []
for sequence in sequences:
    seq = sequence.split()
    X_sequence = [word_to_index[word] for word in seq[:-1]]
    X.append(X_sequence)
    y.append(word_to_index[seq[-1]])
# 将x和y转换为Numpy数组
X = np.array(X)
y = np_utils.to_categorical(y, num_classes=len(unique_words))
# 创建LSTM模型
model = Sequential()
model.add(LSTM(512, input_shape=(X.shape[1], X.shape[2]), return_sequences=True))
model.add(Dropout(0.2))
model.add(LSTM(512))
model.add(Dropout(0.2))
model.add(Dense(y.shape[1], activation='softmax'))
# 编译模型
model.compile(loss='categorical_crossentropy', optimizer='adam')

# 训练模型
filepath="model_weights_saved.hdf5"
checkpoint = ModelCheckpoint(filepath, monitor='loss', verbose=1, save_best_only=True, mode='min')
model.fit(X, y, epochs=50, batch_size=512, callbacks=[checkpoint])
# 加载已经训练好的模型
filename = "model_weights_saved.hdf5"
model.load_weights(filename)
model.compile(loss='categorical_crossentropy', optimizer='adam')

# 生成文本
from random import randint
from pickle import dump
from keras.models import load_model
from keras.preprocessing.sequence import pad_sequences
# 检索一到多个不同长度序列中的随机序列
def generate_seq(model, mapping, seq_length, seed_text, n_chars):
    in_text = seed_text
    # 结果文本
    results = []
    # 进行文本测试，循环获取字符
    for _ in range(n_chars):
        # 将input字符进行文本处理，转为索引号
        encoded = [mapping[word] for word in in_text.split()]
        # 进行padding，确保到达公共长度
        encoded = pad_sequences([encoded], maxlen=seq_length, truncating='pre')
        # 模型进行测试，预测下一个字符
        yhat = model.predict_classes(encoded, verbose=0)
        # 将文本转化为字符输出
        out_word = ''
        for word, index in mapping.items():
            if index == yhat:
                out_word = word
                break
        # 添加结果
        in_text += ' ' + out_word
        results.append(out_word)
    return ' '.join(results)
# 选择一个随机序列
seed_text = sequences[randint(0,len(sequences))]
print(seed_text + '\n')
# 用模型生成新的文本
generated = generate_seq(model, word_to_index, sequence_length, seed_text, 50)
print(generated)