keras cnn以及lstm测试

最新推荐文章于 2025-06-24 12:57:14 发布

HxShine

最新推荐文章于 2025-06-24 12:57:14 发布

阅读量2.1k

点赞数 2

CC 4.0 BY-SA版权

分类专栏：深度学习 keras

本文链接：https://blog.csdn.net/qq_16949707/article/details/84973051

深度学习同时被 2 个专栏收录

28 篇文章

订阅专栏

keras

4 篇文章

订阅专栏

本文详细介绍并实践了三种深度学习模型：卷积神经网络（CNN）、长短时记忆网络（LSTM）及CNN与LSTM的结合使用。通过创建虚假数据集，演示了如何使用Keras库构建、训练和评估这些模型，为读者提供了深入理解模型工作原理的机会。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

1.cnn测试

import numpy as np
import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras.optimizers import SGD

# 创建虚假数据
x_train = np.random.random((100, 100, 100, 3))
y_train = keras.utils.to_categorical(np.random.randint(10, size=(100, 1)), num_classes=10)
x_test = np.random.random((20, 100, 100, 3))
y_test = keras.utils.to_categorical(np.random.randint(10, size=(20, 1)), num_classes=10)
model = Sequential()
# 输入: 100x100的3通道图像 -> 张量 (100, 100, 3).
# 在每块3x3的区域应用32个卷积过滤器.
model.add(Conv2D(32, (3, 3), activation='relu', input_shape=(100, 100, 3)))
model.add(Conv2D(32, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))

sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='categorical_crossentropy', optimizer=sgd)

model.fit(x_train, y_train, batch_size=2, epochs=1)

参考：
http://xiaosheng.me/2017/07/04/article77/

2.lstm测试

from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.layers import Embedding
from keras.layers import LSTM
import keras
import numpy as np

data_dim = 10 #词特征维度
timesteps = 7 #句子长度
num_classes = 9 #类别数
max_features=8 #词汇量
batch=5

# 创建虚假的训练数据
x_train = np.random.randint(max_features,size=(batch, timesteps)) #存的是各个字在embedding中的index
y_train = np.random.randint(num_classes, size=(batch, 1))

model = Sequential()
model.add(Embedding(max_features, output_dim=20, input_length=timesteps))
model.add(LSTM(20))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))

model.compile(loss='binary_crossentropy',
              optimizer='rmsprop',
              metrics=['accuracy'])

model.fit(x_train, y_train, batch_size=16, epochs=1)

参考：
http://xiaosheng.me/2017/07/04/article77/
http://frankchen.xyz/2017/12/18/How-to-Use-Word-Embedding-Layers-for-Deep-Learning-with-Keras/

3.cnn 联合 lstm测试

3.1 输入维度

embedding_con.shape,embed_size,max_features,maxlen
#((100000, 1200), 300, 100000, 40)

3.2 测试代码

    inp = Input(shape=(maxlen,))
    embed = Embedding(max_features, embed_size*4,weights=[embedding_con], trainable=False)(inp)
    filter_sizes = [1, 2, 3, 5]
    num_filters = 300
    conv_0 = Conv1D(num_filters, filter_sizes[0], padding='same', kernel_initializer='normal', activation='relu')(
        embed)
    conv_1 = Conv1D(num_filters, filter_sizes[1], padding='same', kernel_initializer='normal', activation='relu')(
        embed)
    conv_2 = Conv1D(num_filters, filter_sizes[2], padding='same', kernel_initializer='normal', activation='relu')(
        embed)
    conv_3 = Conv1D(num_filters, filter_sizes[3], padding='same', kernel_initializer='normal', activation='relu')(
        embed)
    l_0 = Bidirectional(CuDNNLSTM(40, return_sequences=True))(conv_0)
    l_1 = Bidirectional(CuDNNLSTM(40, return_sequences=True))(conv_1)
    l_2 = Bidirectional(CuDNNLSTM(40, return_sequences=True))(conv_2)
    l_3 = Bidirectional(CuDNNLSTM(40, return_sequences=True))(conv_3)
    
    atten_0 = Attention(maxlen)(l_0)
    atten_1 = Attention(maxlen)(l_1)
    atten_2 = Attention(maxlen)(l_2)
    atten_3 = Attention(maxlen)(l_3)
    
    conc = concatenate([atten_0, atten_1, atten_2, atten_3])
    d1 = Dense(16, activation="relu")(conc)
    dp1 = Dropout(0.1)(d1)
    outp = Dense(1, activation="sigmoid")(dp1)
    print('embed.shape:',embed.shape)
    print('conv_0.shape:',conv_0.shape)
    print('conv_1.shape:',conv_1.shape)
    print('conv_2.shape:',conv_2.shape)
    print('conv_3.shape:',conv_3.shape)
    print('l_0.shape:',l_0.shape)
    print('l_1.shape:',l_1.shape)
    print('l_2.shape:',l_2.shape)
    print('l_3.shape:',l_3.shape)
    print('atten_0.shape:',atten_0.shape)
    print('atten_1.shape:',atten_1.shape)
    print('atten_2.shape:',atten_2.shape)
    print('atten_3.shape:',atten_3.shape)
    print('conc.shape:',conc.shape)
    print('d1.shape:',d1.shape)
    print('dp1.shape:',dp1.shape)
    print('outp.shape:',outp.shape)

3.3 输出

embed.shape: (?, 40, 1200)
conv_0.shape: (?, 40, 300)
conv_1.shape: (?, 40, 300)
conv_2.shape: (?, 40, 300)
conv_3.shape: (?, 40, 300)
l_0.shape: (?, 40, 80)
l_1.shape: (?, 40, 80)
l_2.shape: (?, 40, 80)
l_3.shape: (?, 40, 80)
atten_0.shape: (?, 80)
atten_1.shape: (?, 80)
atten_2.shape: (?, 80)
atten_3.shape: (?, 80)
conc.shape: (?, 320)
d1.shape: (?, 16)
dp1.shape: (?, 16)
outp.shape: (?, 1)