keras api：
https://keras.io/callbacks/

源码：
https://github.com/keras-team/keras/blob/master/keras/callbacks/callbacks.py#L275

内置回调函数
https://blog.csdn.net/wf592523813/article/details/95027246?depth_1-utm_source=distribute.pc_relevant.none-task&utm_source=distribute.pc_relevant.none-task

自定义回调函数
https://blog.csdn.net/qq_27825451/article/details/93377801?depth_1-utm_source=distribute.pc_relevant.none-task&utm_source=distribute.pc_relevant.none-task

例子：
自定义回调函数，打印每个batch的信息，每个epoch的信息

class PrintInfoCallback(tf.keras.callbacks.Callback):  # 继承自Callback类


    # 在每一个batch结束后记录相应的值
    def on_batch_end(self, batch, logs=None):
#         print("logs::::",logs)
        print("batch ===== loss:",logs.get('loss'),"acc:",logs.get('acc'),"val_loss:",logs.get('loss'),"val_acc:",logs.get('val_acc'))

    # 在每一个epoch之后记录相应的值
    def on_epoch_end(self, epoch, logs=None):
#         会接着后面打印，所以开始处加上了换行符
        print("\nepoch ===== loss:",logs.get('loss'),"acc:",logs.get('acc'),"val_loss:",logs.get('loss'),"val_acc:",logs.get('val_acc'))

使用

pic = PrintInfoCallback() 

history=model.fit(
    train_images,
    train_labels,
    epochs=5,
    validation_data=(test_images,test_labels),callbacks=[pic])

原文出处：/show_1EF537kY4Q1N.html