Keras 对序列进行一维和二维卷积

摘要：网络结构来自固定随机数种子以复现结果创建维向量，并扩展维度适应对输入的要求，的大小为定义卷积层卷积核数量为卷积核大小为定义最大化池化层平铺层，调整维度适应全链接层定义全链接层编译模型打印层的输出打印网络结构最终输出如下卷积结果网络结

网络结构来自https://github.com/nfmcclure/...

import numpy as np
import keras

# 固定随机数种子以复现结果
seed=13
np.random.seed(seed)

# 创建 1 维向量，并扩展维度适应 Keras 对输入的要求， data_1d 的大小为 (1, 25, 1)
data_1d = np.random.normal(size=25)
data_1d = np.expand_dims(data_1d, 0)
data_1d = np.expand_dims(data_1d, 2)

# 定义卷积层
filters = 1 # 卷积核数量为 1
kernel_size = 5 # 卷积核大小为 5
convolution_1d_layer = keras.layers.convolutional.Conv1D(filters, kernel_size, strides=1, padding="valid", input_shape=(25, 1), activation="relu", name="convolution_1d_layer")

# 定义最大化池化层
max_pooling_layer = keras.layers.MaxPool1D(pool_size=5, strides=1, padding="valid", name="max_pooling_layer")

# 平铺层，调整维度适应全链接层
reshape_layer = keras.layers.core.Flatten(name="reshape_layer")

# 定义全链接层
full_connect_layer = keras.layers.Dense(5, kernel_initializer=keras.initializers.RandomNormal(mean=0.0, stddev=0.1, seed=seed), bias_initializer="random_normal", use_bias=True, name="full_connect_layer")

# 编译模型
model = keras.Sequential()
model.add(convolution_1d_layer)
model.add(max_pooling_layer)
model.add(reshape_layer)
model.add(full_connect_layer)

# 打印 full_connect_layer 层的输出
output = keras.Model(inputs=model.input, outputs=model.get_layer("full_connect_layer").output).predict(data_1d)
print(output)

# 打印网络结构
print(model.summary())

最终输出如下

======================卷积结果=========================
[[-0.0131043  -0.11734447  0.13395447 -0.75453871 -0.69782442]]
======================网络结构=========================
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
convolution_1d_layer (Conv1D (None, 21, 1)             6         
_________________________________________________________________
max_pooling_layer (MaxPoolin (None, 17, 1)             0         
_________________________________________________________________
reshape_layer (Flatten)      (None, 17)                0         
_________________________________________________________________
full_connect_layer (Dense)   (None, 5)                 90        
=================================================================
Total params: 96
Trainable params: 96
Non-trainable params: 0
_________________________________________________________________
None

data_size = [10, 10]
data_2d = np.random.normal(size=data_size)
data_2d = np.expand_dims(data_2d, 0)
data_2d = np.expand_dims(data_2d, 3)
print data_2d.shape

# 定义卷积层
conv_size = 2
conv_stride_size = 2
convolution_2d_layer = keras.layers.Conv2D(filters=1, kernel_size=(conv_size, conv_size), strides=(conv_stride_size, conv_stride_size), input_shape=(data_size[0], data_size[0], 1))
# convolution_2d_layer = keras.layers.Conv2D(filter=1, kernel_size=kernel, strides=[1,1], padding="valid", activation="relu", name="convolution_2d_layer", input_shape=(1, data_size[0], data_size[0]))


# 定义最大化池化层
pooling_size = (2, 2)
max_pooling_2d_layer = keras.layers.MaxPool2D(pool_size=pooling_size, strides=1, padding="valid", name="max_pooling_2d_layer")

# 平铺层，调整维度适应全链接层
reshape_layer = keras.layers.core.Flatten(name="reshape_layer")

# 定义全链接层
full_connect_layer = keras.layers.Dense(5, kernel_initializer=keras.initializers.RandomNormal(mean=0.0, stddev=0.1, seed=seed), bias_initializer="random_normal", use_bias=True, name="full_connect_layer")

model_2d = keras.Sequential()
model_2d.add(convolution_2d_layer)
model_2d.add(max_pooling_2d_layer)
model_2d.add(reshape_layer)
model_2d.add(full_connect_layer)

# 打印 full_connect_layer 层的输出
output = keras.Model(inputs=model_2d.input, outputs=model_2d.get_layer("full_connect_layer").output).predict(data_2d)
print("======================卷积结果=========================")
print(output)

# 打印网络结构
print("======================网络结构=========================")
print(model_2d.summary())

输出

======================卷积结果=========================
[[ 0.30173036 -0.10435719 -0.03354734  0.24000235 -0.09962128]]
======================网络结构=========================
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 5, 5, 1)           5         
_________________________________________________________________
max_pooling_2d_layer (MaxPoo (None, 4, 4, 1)           0         
_________________________________________________________________
reshape_layer (Flatten)      (None, 16)                0         
_________________________________________________________________
full_connect_layer (Dense)   (None, 5)                 85        
=================================================================
Total params: 90
Trainable params: 90
Non-trainable params: 0
_________________________________________________________________
None