How to record val_loss and loss per batch in keras

Question

I'm using the callback function in keras to record the loss and val_loss per epoch, But I would like to a do the same but per batch. I found a callback function called on_batch_begin(self,batch,log={}), but I not sure how to use it.

Answer 1

Here is an example of custom callback. Following and modifying an example from here:

class LossHistory(keras.callbacks.Callback):
    def on_train_begin(self, logs={}):
        self.losses = []
        self.val_losses = []

    def on_batch_end(self, batch, logs={}):
        self.losses.append(logs.get('loss'))
        self.val_losses.append(logs.get('val_loss'))

model = Sequential()
model.add(Dense(10, input_dim=784, init='uniform'))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='rmsprop')

history = LossHistory()
model.fit(X_train, Y_train, batch_size=128, nb_epoch=20, verbose=0, validation_split=0.1,
          callbacks=[history])

print history.losses
# outputs
'''
[0.66047596406559383, 0.3547245744908703, ..., 0.25953155204159617, 0.25901699725311789]
'''
print history.val_losses