-2

I don't know how interpretation charts from training network in Machine Learning. I don't understand why val_accuracy is divergent compering to accuracy and the same problem in second chart (loss vs val_loss). Please help me. Thanks

charts

I try training in difference optimizer (RMSprop, AdaMax, AdaDelta).

model = Sequential()
model.add(conv_base)
model.add(layers.Flatten())
model.add(layers.Dense(units=512, activation='relu'))
model.add(layers.Dense(units=512, activation='relu'))
model.add(layers.Dense(units=3, activation='softmax'))


model.compile(optimizer=optimizers.RMSprop(learning_rate=0.000001),
             loss='categorical_crossentropy',
             metrics=['accuracy'])

model.summary()


from panel.io import callbacks

callback_list = EarlyStopping(monitor='accuracy', mode='max', patience=7)

history = model.fit_generator(generator=train_generator,
                             steps_per_epoch=steps_per_epoch,
                             epochs=40,    # 100
                             validation_data=valid_generator,
                             validation_steps=validations_steps, callbacks = [callback_list])
Miron Smętek
  • 1
  • 1

1 Answers1

-1

It seems like your are facing a common problem in machine learning called overfitting. Meaning your model is performing better on training data then on validation set of the data. It also may be a thing when your training and validation sets of data aren’t properly shuffled so one of them is more complex for the model to predict.

fedden
  • 319
  • 2
  • 11