# increase lr
lr *= self.lr_mult
K.set_value(self.model.optimizer.lr, lr)
def plot(self):
plt.plot(self.lrs, self.losses)
plt.xscale("log")
plt.xlabel("learning rate (log scale)")
plt.ylabel("loss")
if __name__ == "__main__":
# set up data
x_train, y_train = load_data("train")
num_classes = 12
y_train = keras.utils.to_categorical(y_train, num_classes)
x_train = preprocess_input(x_train)
# set up model
model = DeepYeast()
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.SGD())
# search for lr
lr_finder = LearningRateFinder(model)
lr_finder.find()
lr_finder.plot()
# set up data
x_val, y_val = load_data("val")
x_train, y_train = load_data("train")
num_classes = 12
y_val = keras.utils.to_categorical(y_val, num_classes)
y_train = keras.utils.to_categorical(y_train, num_classes)
x_train = preprocess_input(x_train)
x_val = preprocess_input(x_val)
# set up model
model = DeepYeast()
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
filepath = "../weights-{epoch:02d}-{val_acc:.3f}.hdf5"
checkpoint = keras.callbacks.ModelCheckpoint(filepath,
monitor='val_acc',
verbose=1,
save_best_only=True,
mode='max')
reduce_lr = keras.callbacks.ReduceLROnPlateau(monitor='val_acc',
factor=0.1,
patience=5,
cooldown=0,
min_lr=1e-5)
callbacks_list = [checkpoint, reduce_lr]