dropout_rate = 0.2
initializer = 'he_normal'
weight_decay = 5e-4
regularizer = l2(weight_decay)
# training parameters
epochs = 200
batch_size = 32
learning_rate = 0.01
max_learning_rate = 0.1
clr = OneCycleLR(num_samples=X_train.shape[0],
batch_size=batch_size,
max_lr=max_learning_rate)
chk = ModelCheckpoint(filepath='results/wrn1028',
save_weights_only=True,
monitor='val_loss',
mode='min',
save_best_only=True)
# fit the model
model = WideResNet(width, depth, classes, filters, input_shape, activation,
dropout_rate, initializer, regularizer).get_model()
model.compile(optimizer=SGD(lr=learning_rate),
loss='categorical_crossentropy',
metrics=['accuracy'])
model.fit(generator.flow(X_train, Y_train, batch_size=batch_size),
epochs=epochs,
batch_size=batch_size,
verbose=2,
validation_data=(X_test, Y_test),
callbacks=[clr, chk])
class TrainerV2(object):
def __init__(self,
input_shape,
encode_dim,
output_dim,
model='efficient_net',
loss='emd'):
self.model = None
if model == 'efficient_net':
self.model = EfficientNet(input_shape, encode_dim, output_dim)
elif model == 'wide_res_net':
self.model = WideResNet(input_shape, output_dim)
else:
raise Exception('no match model name')
optimizer = tf.keras.optimizers.SGD(learning_rate=0.1, momentum=0.1)
loss_func = None
if loss == 'emd':
loss_func = EMD
elif loss == 'categorical_crossentropy':
loss_func = 'categorical_crossentropy'
else:
raise Exception('no match loss function')
self.model.compile(optimizer=optimizer,
loss=loss_func,
metrics=['acc'])
def train(self, x_train, t_train, x_val, t_val, epochs, batch_size,
image_path, save_name):
train_gen = DataGenerator(x_train,
t_train,
image_path=image_path,
batch_size=batch_size)
val_gen = DataGenerator(x_val,
t_val,
image_path=image_path,
batch_size=batch_size)
callbacks = [
tf.keras.callbacks.ModelCheckpoint(save_name,
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min')
]
self.history = self.model.fit_generator(
train_gen,
len(train_gen),
epochs=30,
validation_data=val_gen,
validation_steps=len(val_gen),
callbacks=callbacks,
)
def evaluate(
self,
x_test,
t_test,
batch_size,
image_path,
):
test_gen = DataGenerator(x_test,
t_test,
image_path=image_path,
batch_size=batch_size)
preds = self.model.predict_generator(
test_gen,
len(test_gen),
)
idx = np.array([0, 1, 2, 3, 4])
acc1 = accuracy_score(np.argmax(t_test, axis=1),
np.argmax(preds, axis=1))
acc2 = accuracy_score(np.argmax(t_test, axis=1),
np.sum(preds * idx, axis=1).astype(np.int32))
cm = confusion_matrix(np.argmax(t_test, axis=1),
np.argmax(preds, axis=1))
print(acc1, acc2, cm)
return (acc1, acc2, cm)