Using multiple callbacks at once..

[Noob]

I am currently training my network using crossvalidation, but have no idea how many epochs i need to train it for. 

I see that you can append callbacks for the model training.  Where the one called ReduceLROnPlateau and EarlyStopping seems interesting. But can these be used at the same time?

seed = 7
np.random.seed(seed)
kfold = KFold(n_splits=10, shuffle=False, random_state=None)
print "Splits"
cvscores_acc = []
cvscores_loss = []
hist = []
i = 0
for train, test in kfold.split(train_set_data_vstacked_normalized):


    print "Model definition!"
    model = Sequential()


    #act = PReLU(init='normal', weights=None)
    model.add(Dense(output_dim=400,input_dim=400, init="normal",activation=K.tanh))


    #act1 = PReLU(init='normal', weights=None)
    model.add(Dense(output_dim=400,input_dim=400, init="normal",activation=K.tanh))


    #act2 = PReLU(init='normal', weights=None)
    #model.add(Dense(output_dim=400, input_dim=400, init="normal",activation=K.tanh))


    act4=ELU(100)
    model.add(Dense(output_dim=13, input_dim=400, init="normal",activation=act4))


    print "Compiling"
    model.compile(loss='mean_squared_error', optimizer='RMSprop')
    print "Compile done! "


    print '\n'


    print "Train start"


    reduce_lr=ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=10, verbose=1, mode='auto', epsilon=0.0001, cooldown=0, min_lr=0)
    log=csv_logger = CSVLogger('training_'+str(i)+'.csv')
    #early_stop=EarlyStopping(monitor='val_loss', min_delta=0, patience=0, verbose=1, mode='auto')




    hist_current = model.fit(train_set_data_vstacked_normalized[test],train_set_output_vstacked[test], shuffle=False,validation_split=0.1 ,  nb_epoch=1000,verbose=1,callbacks=[reduce_lr,log])


    hist.append(hist_current)


    loss, accuracy = model.evaluate(x=train_set_data_vstacked_normalized[train],y=train_set_output_vstacked[train],verbose=1)
    print
    print('loss: ', loss)
    print('accuracy: ', accuracy)
    print()
    print model.summary()
    print "New Model:"
    cvscores_acc.append(accuracy)
    cvscores_loss.append(loss)
    print "Model stored"
    model.save("Model"+str(i))
    i=i+1


print("%.2f%% (+/- %.2f%%)" % (numpy.mean(cvscores_acc), numpy.std(cvscores_acc)))
print("%.2f%% (+/- %.2f%%)" % (numpy.mean(cvscores_loss), numpy.std(cvscores_loss)))

So i use earlystopping to stop the training if the val_los don't change after a while and  ReduceLROnPlateau reduce the learning rate when no decrease in val_loss has been seen. 
I tried it, but earlystopping didn't seem to be very fond of the model fit in the for loop.

[acha...@gmail.com]

I had a similar problem. I fixed it by doing something like this:

    earlyStopImprovement = EarlyStopping(monitor='loss', min_delta=0, patience=10, verbose=verbose, mode='auto')

    checkpoint = ModelCheckpoint(filepath, monitor='loss', verbose=verbose, save_best_only=True, mode='auto')

    callbacks_list = [checkpoint,earlyStopImprovement]

    model.fit(X1, Y1, epochs=5000, batch_size=10,  verbose=2, callbacks=callbacks_list)

Note that in the fit call, it's a single variable, not a list.

[Loser]

As for solution I just added in stop in the callback list and everything worked.