##每周學(xué)一點(diǎn)--Python深度學(xué)習(xí)篇

##基于python和tensorflow的網(wǎng)絡(luò)微調(diào)（fine tunning） P117

1.在已訓(xùn)練好的base network上添加自定義網(wǎng)絡(luò)，

Conv2D, MaxPooling2D... + Flatten, Dense(自己添加的網(wǎng)絡(luò)，用于訓(xùn)練)

code:

model = models.Sequential()

model.add(conv_base) #訓(xùn)練好的卷積、池化網(wǎng)絡(luò)層

model.add(layers.Flatten()) #將上層的輸出展平到一維

model.add(layers.Dense(256, activation='relu'))

model.add(layers.Dense(1, activation='sigmoid')) #最后變成單一輸出，輸出值位0~1概率

2.凍結(jié)base network

code:

conv_base.trainable = False #將base network可傳播屬性設(shè)為false

3.訓(xùn)練所添加網(wǎng)絡(luò)部分

code:

history = model.fit_generator(

train_generator,

steps_per_epoch=100,

epochs=30,

validation_data=validation_generator,

validation_steps=50)

4.解凍一些base network(從后往前)

通過設(shè)置參數(shù)layer.trainable = True

5.重新訓(xùn)練整個(gè)網(wǎng)絡(luò)

#模型結(jié)構(gòu)查看

model.summary()

##卷積神經(jīng)網(wǎng)絡(luò)可視化

#關(guān)于numpy.expand_dims(a, axis), 在axis參數(shù)位置添加一個(gè)維度？

import numpy as np

test = np.array([[1, 2, 3], [4, 5, 6]])

np.shape(test): (2, 3)

x = np.expand_dims(test, axis=0)

np.shape(x): (1, 2, 3)

y = np.expand_dims(test, axis=1)

np.shape(y): (2, 1, 3)

z = np.expand_dims(test, axis=2)

np.shape(z): (2, 3, 1)