TensorFlow,keras打印中间某一层结果

在大数据建模过程中&＃xff0c;希望融合多种特征&＃xff0c;因此需要将模型中间的是特征保存下来&＃xff0c;以便后面使用。

1.定义模型

def arch(data):Random(0)model &＃61; keras.Sequential([layers.Input(shape&＃61;(data.shape[1], data.shape[2])),layers.Conv1D(filters&＃61;32, kernel_size&＃61;7, padding&＃61;"same", strides&＃61;2, activation&＃61;"relu"),layers.Dropout(rate&＃61;0.2),layers.Conv1D(filters&＃61;16, kernel_size&＃61;7, padding&＃61;"same", strides&＃61;2, activation&＃61;"relu"),layers.Conv1DTranspose(filters&＃61;16, kernel_size&＃61;7, padding&＃61;"same", strides&＃61;2, activation&＃61;"relu"),layers.Dropout(rate&＃61;0.2),layers.Conv1DTranspose(filters&＃61;32, kernel_size&＃61;7, padding&＃61;"same", strides&＃61;2, activation&＃61;"relu"),layers.Conv1DTranspose(filters&＃61;1, kernel_size&＃61;7, padding&＃61;"same"),])model.compile(optimizer&＃61;keras.optimizers.Adam(learning_rate&＃61;0.001), loss&＃61;"mse")model.summary()history &＃61; model.fit(data,data,epochs&＃61;100,batch_size&＃61;32,validation_split&＃61;0.1,verbose&＃61;0,callbacks&＃61;[keras.callbacks.EarlyStopping(monitor&＃61;"val_loss", patience&＃61;5, mode&＃61;"min", verbose&＃61;0)],)return history, model


查看模型架构

model.summary()


2.获取conv1d_transpose_4层的输出结果。代码如下&＃xff1a;

from keras import backend as K
representation_layer &＃61; K.function(inputs&＃61;[model.layers[0].input], outputs&＃61;[model.get_layer(&＃39;conv1d_transpose_4&＃39;).output])representation &＃61; representation_layer([X])
representation &＃61; np.array(representation)[0]
print(representation.shape)
print(type(representation))
print(representation)


结果如图&＃xff0c;保存即可。

参考链接&＃xff1a;https://blog.csdn.net/selectopti/article/details/115933551