Keras ModelCheckpoint:保存_freq/周期可以动态变化吗？开发者社区

link管理

链接快照平台

输入网页链接，自动生成快照
标签化管理网页链接

文章/答案/技术大牛

发布

# [...]
batch_size = 32
epochs0    = 10
epochs     = 110
callbacks0 = [
    keras.callbacks.ModelCheckpoint(
        resuts_dir+'model0/',
        monitor='val_loss',
        verbose=1,
        save_best_only=False,
        save_weights_only=False,
        save_freq=batch_size*epochs0)
callbacks  = [
    keras.callbacks.ModelCheckpoint(
        resuts_dir+'model/',
        monitor='val_loss',
        verbose=1,
        save_best_only=True,
        save_weights_only=False,
        save_freq='epoch'),
    keras.callbacks.EarlyStopping(
        monitor='val_loss',
        min_delta=0,
        patience=20,
        verbose=1,
        mode='auto')
# [...]
training_history0 = model.fit(train_gen, epochs=epochs0, validation_data=val_gen, callbacks=callbacks0, verbose=1)
training_history  = model.fit(train_gen, epochs=epochs,  validation_data=val_gen, callbacks=callbacks,  verbose=1)

class CustomCheckpoint(tf.keras.callbacks.ModelCheckpoint):
    def __init__(self, filepath, monitor, verbose, save_best_only, 
                 save_weights_only, min_epochs, save_freq):
        self.min_epochs = min_epochs
        super().__init__(filepath=filepath, 
                         monitor=monitor,
                         verbose=verbose,
                         save_best_only=save_best_only,
                         save_weights_only=save_weights_only
                         save_freq=save_freq)
    def on_epoch_end(self, epoch, logs={}):
        if self.save_freq == 'epoch':
            if (epoch+1) >= self.min_epochs:
                self._save_model(epoch=epoch, logs=logs)

MIN_EPOCHS = 10
custom_callback = CustomCheckpoint(filepath="checkpoint_e{epoch:02d}", # this filepath includes the epoch name
                                   min_epochs=MIN_EPOCHS, 
                                   monitor='val_loss',

问 Keras ModelCheckpoint:保存_freq/周期可以动态变化吗？ EN

回答 1

Stack Overflow用户