keras和tensorfboard的结合
keras通过结合tensorfboard,可以很直观的分析迭代过程中的指标变化,为研究指明方向
1.通过回调函数,保存训练记录
from keras.callbacks import TensorBoard
from keras.models import Sequential
finetuned_model = Model(model.input, x)
finetuned_model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
finetuned_model.fit_generator(batches, steps_per_epoch=num_train_steps, epochs=50, callbacks=[early_stopping, checkpointer,history,TensorBoard(log_dir='./tmp/log')], validation_data=val_batches, validation_steps=num_valid_steps)
对,就是callbacks参数中的TensorBoard函数,里面装的是log存储路径
2.调用tensorboard
让后再cmd/terminal下命令: tensorboard --logdir=./temp/log 来浏览器进行查看
注意,这里的路径一定要写绝对路径