Tracking learning

Pemodelan Citra dengan Keras

Ariel Rokem

Senior Data Scientist, University of Washington

Learning curves: training

Learning curves: validation

Learning curves: overfitting

training = model.fit(train_data, train_labels, 
                     epochs=3, validation_split=0.2)

import matplotlib.pyplot as plt
plt.plot(training.history['loss'])

plt.plot(training.history['val_loss'])

plt.show()

Storing the optimal parameters

from keras.callbacks import ModelCheckpoint

# This checkpoint object will store the model parameters 
# in the file "weights.hdf5"
checkpoint = ModelCheckpoint('weights.hdf5', monitor='val_loss', 
                             save_best_only=True)

# Store in a list to be used during training
callbacks_list = [checkpoint]

# Fit the model on a training set, using the checkpoint as a 
#callback
model.fit(train_data, train_labels, validation_split=0.2, 
          epochs=3, callbacks=callbacks_list)

Loading stored parameters

model.load_weights('weights.hdf5')

model.predict_classes(test_data)

array([2, 2, 1, 2, 0, 1, 0, 1, 2, 0])

Let's practice!

Pemodelan Citra dengan Keras