def fit_model():
start = 0.03
stop = 0.001
nb_epoch = 10000
PRETRAIN = False
learning_rate = np.linspace(start, stop, nb_epoch)
X, y = load2d()
X_train, X_test, y_train, y_test = train_test_split(X, y,
test_size=0.2, random_state=42)
model = cnn_model()
if PRETRAIN:
model.load_weights('my_cnn_model_weights.h5')
sgd = SGD(lr=start, momentum=0.9, nesterov=True)
model.compile(loss='mse', optimizer=sgd)
change_lr = LearningRateScheduler(lambda epoch: float(learning_rate[epoch]))
early_stop = EarlyStopping(patience=100)
flipgen = FlippedImageDataGenerator()
hist = model.fit_generator(flipgen.flow(X_train, y_train),
samples_per_epoch=X_train.shape[0],
nb_epoch=nb_epoch,
validation_data=(X_test, y_test),
callbacks=[change_lr, early_stop])
model.save_weights('my_cnn_model_weights.h5', overwrite=True)
np.savetxt('my_cnn_model_loss.csv', hist.history['loss'])
np.savetxt('my_cnn_model_val_loss.csv', hist.history['val_loss'])
def fit_specialists():
specialists = OrderedDict()
start = 0.03
stop = 0.001
nb_epoch = 1
PRETRAIN = False
learning_rate = np.linspace(start, stop, nb_epoch)
for setting in SPECIALIST_SETTINGS:
cols = setting['columns']
X, y = load2d(cols=cols)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.2,
random_state=42)
model = cnnn_model()
if PRETRAIN:
model.load_weights('my_cnn_model_weights.h5')
model.layers.pop()
model.outputs = [model.layers[-1].output]
model.layers[-1].outbound_nodes = []
model.add(Dense(len(cols)))
sgd = SGD(lr=start, momentum=0.9, nesterov=True)
model.compile(loss='mse', optimizer=sgd)
lr_decay = LearningRateScheduler(
lambda epoch: float(learning_rate[epoch]))
early_stop = EarlyStopping(patience=100)
flipgen = FlippedImageDataGenerator()
flipgen.flip_indices = setting['flip_indices']
print('Training model for columns {} for {} epochs'.format(
cols, nb_epoch))
hist = model.fit_generator(flipgen.flow(X_train, y_train),
samples_per_epoch=X_train.shape[0],
nb_epoch=nb_epoch,
validation_data=(X_test, y_test),
callbacks=[lr_decay, early_stop])
model.save_weights('my_cnn_model_{}_weights.h5'.format(cols[0]))
np.savetxt('my_cnn_model_{}_loss.csv'.format(cols[0]),
hist.history['loss'])
np.savetxt('my_cnn_model_{}_val_loss.csv'.format(cols[0]),
hist.history['val_loss'])
specialists[cols] = model
def fit_specialists():
specialists = OrderedDict()
start = 0.03
stop = 0.001
nb_epoch = 10000
PRETRAIN = False
learning_rate = np.linspace(start, stop, nb_epoch)
for setting in SPECIALIST_SETTINGS:
cols = setting['columns']
X, y = load2d(cols=cols)
X_train, X_test, y_train, y_test = train_test_split(X, y,
test_size=0.2, random_state=42)
model = cnn_model()
if PRETRAIN:
model.load_weights('my_cnn_model_weights.h5')
model.layers.pop()
model.outputs = [model.layers[-1].output]
model.layers[-1].outbound_nodes = []
model.add(Dense(len(cols)))
sgd = SGD(lr=start, momentum=0.9, nesterov=True)
model.compile(loss='mse', optimizer=sgd)
lr_decay = LearningRateScheduler(lambda epoch: float(learning_rate[epoch]))
early_stop = EarlyStopping(patience=100)
flipgen = FlippedImageDataGenerator()
flipgen.flip_indices = setting['flip_indices']
print('Training model for columns {} for {} epochs'.format(cols, nb_epoch))
hist = model.fit_generator(flipgen.flow(X_train, y_train),
samples_per_epoch=X_train.shape[0],
nb_epoch=nb_epoch,
validation_data=(X_test, y_test),
callbacks=[lr_decay, early_stop])
model.save_weights('my_cnn_model_{}_weights.h5'.format(cols[0]))
np.savetxt('my_cnn_model_{}_loss.csv'.format(cols[0]), hist.history['loss'])
np.savetxt('my_cnn_model_{}_val_loss.csv'.format(cols[0]), hist.history['val_loss'])
specialists[cols] = model