版权声明：本文为博主原创文章，转载请注明出处。 https://blog.csdn.net/Yellow_python/article/details/86707663
内置样本
影评正负2分类（已编码）
from keras.datasets import imdb # Internet Movie Database
# num_words设定较小时，会发现高频词多是停词
(x, y), _ = imdb.load_data(num_words=1)
print(x.shape, y.shape) # (25000,) (25000,)
# 词与ID间的映射
word2id = imdb.get_word_index()
id2word = {i: w for w, i in word2id.items()}
print(x[0])
print(' '.join([id2word[i] for i in x[0]]))
序列截断或补齐为等长
from keras.preprocessing.sequence import pad_sequences
maxlen = 2
print(pad_sequences([[1, 2, 3], [1]], maxlen))
"""[[2 3] [0 1]]"""
print(pad_sequences([[1, 2, 3], [1]], maxlen, value=9))
"""[[2 3] [9 1]]"""
print(pad_sequences([[1, 2, 3], [1]], maxlen, padding='post'))
"""[[2 3] [1 0]]"""
print(pad_sequences([[1, 2, 3], [1]], maxlen, truncating='post'))
"""[[1 2] [0 1]]"""
词嵌入
from keras.datasets.imdb import load_data # 影评情感2分类
from keras.preprocessing.sequence import pad_sequences
from keras.models import Sequential
from keras.layers import Embedding, Flatten, Dense
"""配置"""
num_words = 10000 # 按词频大小取样本前10000个词
input_dim = num_words # 词库大小（必须>=num_words）
maxlen = 25 # 序列长度
output_dim = 40 # 词向量维度
batch_size = 128
epochs = 2
"""数据读取与处理"""
(x, y), _ = load_data(num_words=num_words)
x = pad_sequences(x, maxlen)
"""建模"""
model = Sequential()
# 词嵌入：词库大小、词向量维度、固定序列长度
model.add(Embedding(input_dim, output_dim, input_length=maxlen))
# 平坦化：maxlen * output_dim
model.add(Flatten())
# 输出层：2分类
model.add(Dense(units=1, activation='sigmoid'))
# RMSprop优化器、二元交叉熵损失
model.compile('rmsprop', 'binary_crossentropy', ['acc'])
# 训练
model.fit(x, y, batch_size, epochs)
"""模型可视化"""
model.summary()
词嵌入+RNN
from keras.datasets.imdb import load_data # 影评情感2分类
from keras.preprocessing.sequence import pad_sequences
from keras.models import Sequential
from keras.layers import Embedding, Dense, SimpleRNN
import matplotlib.pyplot as mp
"""配置"""
num_words = 10000 # 按词频大小取样本前10000个词
input_dim = num_words # 词库大小（必须>=num_words）
maxlen = 25 # 序列长度
output_dim = 40 # 词向量维度
units = 32 # RNN神经元数量
batch_size = 128
epochs = 3
"""数据读取与处理"""
(x, y), _ = load_data(num_words=num_words)
x = pad_sequences(x, maxlen)
"""建模"""
model = Sequential()
model.add(Embedding(input_dim, output_dim, input_length=maxlen))
model.add(SimpleRNN(units, return_sequences=True)) # 返回序列全部结果
model.add(SimpleRNN(units, return_sequences=False)) # 返回序列最尾结果
model.add(Dense(units=1, activation='sigmoid'))
model.summary()
"""编译、训练"""
model.compile('rmsprop', 'binary_crossentropy', ['acc'])
history = model.fit(x, y, batch_size, epochs, verbose=2,
validation_split=.1) # 取10%样本作验证
"""精度曲线"""
acc = history.history['acc']
val_acc = history.history['val_acc']
mp.plot(range(epochs), acc)
mp.plot(range(epochs), val_acc)
mp.show()
词嵌入+CNN
from keras.datasets.imdb import load_data # 影评情感2分类
from keras.preprocessing.sequence import pad_sequences
from keras.models import Sequential
from keras.layers import Embedding, Dense,\
Conv1D, MaxPool1D, GlobalMaxPool1D
"""配置"""
num_words = 10000 # 按词频大小取样本前10000个词
input_dim = num_words # 词库大小（必须>=num_words）
maxlen = 25 # 序列长度
output_dim = 40 # 词向量维度
filters = 32 # 卷积层滤波器数量
kernel_size = 7 # 卷积层滤波器大小
batch_size = 128
epochs = 3
"""数据读取与处理"""
(x, y), _ = load_data(num_words=num_words)
x = pad_sequences(x, maxlen)
"""建模"""
model = Sequential()
model.add(Embedding(input_dim, output_dim, input_length=maxlen))
model.add(Conv1D(filters, kernel_size, activation='relu'))
model.add(MaxPool1D(pool_size=2)) # strides默认等于pool_size
model.add(Conv1D(filters, kernel_size, activation='relu'))
model.add(GlobalMaxPool1D()) # 对于时序数据的全局最大池化
model.add(Dense(units=1, activation='sigmoid'))
model.summary()
"""编译、训练"""
model.compile('rmsprop', 'binary_crossentropy', ['acc'])
model.fit(x, y, batch_size, epochs, 2)
CNN+RNN
from keras.datasets.imdb import load_data # 影评情感2分类
from keras.preprocessing.sequence import pad_sequences
from keras.models import Sequential
from keras.layers import Embedding, Dense,\
Conv1D, MaxPool1D, GRU # Gated Recurrent Unit
"""配置"""
num_words = 10000 # 按词频大小取样本前10000个词
input_dim = num_words # 词库大小（必须>=num_words）
maxlen = 25 # 序列长度
output_dim = 40 # 词向量维度
filters = 32 # 卷积层滤波器数量
kernel_size = 7 # 卷积层滤波器大小
units = 32 # RNN神经元数量
batch_size = 128
epochs = 3
"""数据读取与处理"""
(x, y), _ = load_data(num_words=num_words)
x = pad_sequences(x, maxlen)
"""建模"""
model = Sequential()
model.add(Embedding(input_dim, output_dim, input_length=maxlen))
model.add(Conv1D(filters, kernel_size, activation='relu'))
model.add(MaxPool1D())
model.add(Conv1D(filters, kernel_size, activation='relu'))
model.add(GRU(units)) # 门限循环单元网络
model.add(Dense(1, activation='sigmoid'))
model.summary()
"""编译、训练"""
model.compile('rmsprop', 'binary_crossentropy', ['acc'])
model.fit(x, y, batch_size, epochs, 2)
