def train(model_name):
# load the data
(train_images, train_labels), (test_images,
test_labels) = fashion_mnist.load_data()
# create a TensorBoard callback
model_dir = root_dir(os.path.join('training', model_name))
tensorboard_callback = TensorBoard(log_dir=model_dir)
# create a checkpoint callback
checkpoint_path = root_dir(os.path.join(model_dir, 'weights.ckpt'))
checkpoint_callback = ModelCheckpoint(checkpoint_path,
save_weights_only=True)
# train the model
model = create_model()
model.fit(train_images,
train_labels,
epochs=5,
callbacks=[checkpoint_callback, tensorboard_callback],
validation_data=(test_images, test_labels))
def get_call_back(lr, batch_size):
"""
定义call back
:return:
"""
text = '{}-{}-{}'.format(cfg.base_model_name, batch_size, lr)
checkpoint = ModelCheckpoint(filepath='/tmp/ssd-' + text + '.{epoch:03d}.h5',
monitor='val_loss',
verbose=1,
save_best_only=True,
save_weights_only=True,
save_freq='epoch'
)
reduce = ReduceLROnPlateau(monitor='val_loss', factor=0.2, min_delta=2e-3,
patience=10, min_lr=1e-5)
stop = EarlyStopping(monitor='val_loss', patience=20)
# scheduler = LearningRateScheduler(lr_schedule(epochs, lr))
log = TensorBoard(log_dir='log-{}-{}'.format(text,
time.strftime("%Y%m%d", time.localtime())))
return [checkpoint, reduce, stop, log]
def __init__(self):
self.model_id = None
self.input_param = None
self.nb_steps = None
self.step_length = None
# --- TF ---
self.model = None
self.losses = None
self.loss_lambdas = None
self.opt_param = None
self.decay = None
self.type_loss = None
self.model_options = None
# Spare GPU
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
self.allow_growth()
log_dir = os.path.join('tensorboard', f'{time()}')
self.tensorboard = TensorBoard(log_dir=log_dir)
def train(args, params):
'''
Train model
'''
model = build_model(params)
X_train, Y1_train, Y2_train, X_test, Y1_test, Y2_test = load_dataset(
args.dataset_name, float(args.num_train))
print('Fitting model...')
results = model.fit(
X_train, [Y1_train, Y2_train],
epochs=args.epochs,
verbose=1,
validation_data=(X_test, [Y1_test, Y2_test]),
callbacks=[SendMetrics(),
TensorBoard(log_dir=TENSORBOARD_DIR)])
_, acc = model.evaluate(X_test, [Y1_test, Y2_test], verbose=0)
LOG.debug('Final result is: %d', acc)
nni.report_final_result(acc)
print('Final result is: %d', acc)
def compile_model(self):
# init summary writer for tensorboard
self.callback1 = TensorBoard(self.log_dir + '/discriminator')
self.callback2 = TensorBoard(self.log_dir + '/generator')
self.callback3 = TensorBoard(self.log_dir + '/generated_images')
# model stuff
input_shape = [self.image_size, self.image_size, self.image_channels]
adam1 = Adam(lr=self.lr)
adam2 = Adam(lr=self.lr * 2)
# init and add multi-gpu support
try:
self.discriminator = multi_gpu_model(self.discriminator(),
gpus=self.gpu)
except:
self.discriminator = self.discriminator()
try:
self.generator = multi_gpu_model(self.generator(), gpus=self.gpu)
except:
self.generator = self.generator()
# compile discriminator
self.discriminator.compile(loss='binary_crossentropy', optimizer=adam1)
# compile generator
input_tensor = Input(shape=input_shape)
generated_catroon_tensor = self.generator(input_tensor)
self.discriminator.trainable = False # for here we only train the generator
discriminator_output = self.discriminator(generated_catroon_tensor)
self.train_generator = Model(
input_tensor,
outputs=[generated_catroon_tensor, discriminator_output])
# add multi-gpu support
try:
self.train_generator = multi_gpu_model(self.train_generator,
gpus=self.gpu)
except:
pass
self.train_generator.compile(
loss=[self.vgg_loss, 'binary_crossentropy'],
loss_weights=[float(self.weight), 1.0],
optimizer=adam2)
# set callback model
self.callback1.set_model(self.discriminator)
self.callback2.set_model(self.train_generator)
self.callback3.set_model(self.train_generator)
def createModel(CONV_LAYERS,
CONV_LAYER_SIZE,
DENSE_LAYERS,
DENSE_LAYER_SIZE,
BATCH_SIZE,
OPTIMIZER="adam",
LOSS_FUNCTION="binary_crossentropy",
EPOCHS=2):
lf = LOSS_FUNCTION.replace("_", "-")
global MODEL_NAME
MODEL_NAME = f"R-gyar-prototype-model_convs-{CONV_LAYERS}_convnodes-{CONV_LAYER_SIZE}_denses-{DENSE_LAYERS}_densenodes-{DENSE_LAYER_SIZE}_batch-{BATCH_SIZE}_opt-{OPTIMIZER}_loss-{lf}_epochs-{EPOCHS}_time-{round(time.time())}"
global tensorboard
tensorboard = TensorBoard(log_dir='prototypelogs/{}'.format(MODEL_NAME))
if MODEL_NAME in os.listdir('prototypelogs'):
raise FileExistsError
else:
if Train:
print("Model Created: ", MODEL_NAME)
model = Sequential()
model.add(Conv2D(CONV_LAYER_SIZE, (3, 3), input_shape=(100, 100, 1)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
for l in range(CONV_LAYERS - 1):
model.add(Conv2D(CONV_LAYER_SIZE, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten()
) # this converts our 3D feature maps to 1D feature vectors
for l in range(DENSE_LAYERS):
model.add(Dense(DENSE_LAYER_SIZE))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(1))
model.add(Activation('sigmoid'))
return model
def init_callbacks(self):
experiment = None
if hasattr(self.config, "comet_api_key"):
experiment = Experiment(api_key=self.config.comet_api_key,
project_name=self.config.experiment_name)
experiment.log_parameters(self.config)
self.callbacks.append(experiment.get_keras_callback())
self.callbacks.append(
Evaluater(eval_data=self.data_loader.get_val_generator(),
eval_steps=self.data_loader.get_val_steps(),
ref_data=self.data_loader.get_reference_data(),
config=self.config,
comet_experiment=experiment))
self.callbacks.append(
ReduceLROnPlateau(monitor=self.config.checkpoint_monitor,
factor=self.config.lr_reduce_factor,
patience=self.config.lr_reduce_patience,
verbose=1,
mode=self.config.checkpoint_mode))
self.callbacks.append(
ModelCheckpoint(
filepath=os.path.join(
self.config.checkpoint_dir,
'%s-{epoch:02d}-{val_accuracy:.2f}.hdf5' %
self.config.experiment_name),
monitor=self.config.checkpoint_monitor,
mode=self.config.checkpoint_mode,
save_best_only=self.config.checkpoint_save_best_only,
save_weights_only=self.config.checkpoint_save_weights_only,
verbose=self.config.checkpoint_verbose,
))
self.callbacks.append(
TensorBoard(
log_dir=self.config.tensorboard_log_dir,
write_graph=self.config.tensorboard_write_graph,
))
def train(self,
x_train,
y_train,
learning_rate,
epochs,
batch_size,
tb_logs_dir=None,
verbose=False):
early_stopping_callback = EarlyStopping(monitor="val_loss",
patience=25)
callbacks = [early_stopping_callback]
if bool(tb_logs_dir):
date_time = datetime.now().strftime('%Y-%m-%d-%H%M%S')
log_name = os.path.join(tb_logs_dir,
"{}_{}".format("transfer_net", date_time))
# defining callbacks for training
tensorboard_callback = TensorBoard(log_dir=log_name,
write_graph=True,
write_images=True)
callbacks += [tensorboard_callback]
x_train, x_val, y_train, y_val = train_test_split(x_train,
y_train,
train_size=0.8,
random_state=1)
y_train = tf.keras.utils.to_categorical(y_train,
self.num_output_classes)
y_val = tf.keras.utils.to_categorical(y_val, self.num_output_classes)
self.model.compile(loss="categorical_crossentropy",
optimizer=keras.optimizers.Adam(lr=learning_rate),
metrics=['accuracy', f1_score])
self.hist = self.model.fit(x_train,
y_train,
epochs=epochs,
batch_size=batch_size,
verbose=verbose,
callbacks=callbacks,
validation_data=(x_val, y_val))
def train(self):
dataset = Datasets("./data/brunch")
data, label, mask = dataset.read()
print(data.shape, label.shape, mask.shape)
# save model every 5 epoch
filepath = "./data/checkpoints/init/model-{epoch:02d}.hdf5"
checkpoint = ModelCheckpoint(filepath,
save_weights_only=False,
verbose=1)
# display tensorboard
log_dir = "./data/logs/fit/" + datetime.datetime.now().strftime(
"%Y%m%d-%H%M%S")
tensorboard_callback = TensorBoard(log_dir=log_dir, histogram_freq=1)
self.model.fit(x=[data, label, mask],
y=[mask, mask],
steps_per_epoch=None,
batch_size=self.batch_size,
shuffle=True,
epochs=10,
validation_split=0.1,
callbacks=[checkpoint, tensorboard_callback])
def _build_model(self):
model = Sequential()
# input shape (samples, channels, rows, cols)
model.add(Conv2D(32, (3, 3), strides=(1, 1), input_shape=(24, 10, 1),
kernel_initializer=initializers.glorot_uniform(), activation=activations.relu,
kernel_regularizer=regularizers.l2(0.01))) # kernel initialize weights
model.add(Conv2D(64, (3, 3), strides=(1, 1), activation=activations.relu))
model.add(Conv2D(128, (3, 3), strides=(1, 1), activation=activations.relu))
# model.add(Dropout(0.5))
model.add(Flatten())
model.add(Dense(self.action_size, activation=activations.softmax))
model.compile(loss=losses.categorical_crossentropy, # loss='mse' losses.categorical_crossentropy
optimizer=optimizers.Nadam(lr=self.LEARNING_RATE)) # RMSProb,Adam
self.tensorBoard = TensorBoard('./logs/RLAgent', histogram_freq=0,
write_graph=True, write_images=True)
model.summary()
return model
def __init__(self, gen_file_path, epsilon=1.0):
self.map_gen_file = gen_file_path
self.epsilon = epsilon
self.tiles_per_col = c.COL_HEIGHT - 2 if c.INSERT_GROUND else c.COL_HEIGHT
self.gen_size = c.GEN_LENGTH * self.tiles_per_col
self.checkpoint_gen = os.path.join(dir_path, "checkpoints", "generator")
self._TILE_MAP, self._CHAR_MAP = self.tokenize_tiles(c.GENERATOR_TILES)
# print("generator._TILE_MAP:", self._TILE_MAP)
# print("generator._CHAR_MAP:", self._CHAR_MAP)
self.replay_memory = deque(maxlen=c.REPLAY_MEMORY_SIZE)
loaded_model_path = self.load_model_path() if c.LOAD_GEN_MODEL else None
callback_dir = os.path.join(dir_path, "checkpoints", "generator_graphs", f"graph_{self.start_checkpoint}")
os.makedirs(callback_dir, exist_ok=True)
self.generator = self.create_generator(model=loaded_model_path)
self.tensorboard = TensorBoard(log_dir=callback_dir, histogram_freq=0, write_graph=True, write_images=False)
print("tensorboard callback dir:", callback_dir)
if loaded_model_path:
self.load_replay_memory()
print(self.generator.summary())
def build_checkpointers(self):
checkpointers = []
if self.model_path is not None:
checkpointer = ModelCheckpoint(filepath=self.model_path + '.h5',
monitor='val_loss',
verbose=1,
save_best_only=True,
period=10)
checkpointers.append(checkpointer)
if self.log_dir is not None:
tensorboard = TensorBoard(log_dir=self.log_dir,
histogram_freq=0,
batch_size=32,
write_graph=True,
write_grads=False,
write_images=False)
checkpointers.append(tensorboard)
return checkpointers
def __setup_callbacks(self) -> List:
"""
Sets up the callbacks for training
:return: the early stopping schedule, tensorboard data and the checkpointer
"""
# Create a folder for the model log of the current experiment
weights_log_path = os.path.join(self.__current_experiment_path,
'weights')
# Set up the callback to save the best weights after each epoch
checkpointer = ModelCheckpoint(filepath=os.path.join(
weights_log_path, 'weights.{epoch:02d}-{val_loss:.2f}.hdf5'),
verbose=0,
save_best_only=True,
save_weights_only=True,
monitor='val_loss',
mode='min')
# Set up Tensorboard
tensorboard = TensorBoard(log_dir=os.path.join(
self.__current_experiment_path, 'tensorboard'),
write_graph=True,
histogram_freq=0,
write_grads=True,
write_images=False,
batch_size=self._params.batch_size,
update_freq=self._params.batch_size)
# Set up early stopping to interrupt the training if val_loss is not increasing after n epochs
early_stopping = EarlyStopping(monitor='val_loss',
patience=25,
mode='min')
csv_logger = CSVLogger(os.path.join(self.__current_experiment_path,
"training.csv"),
append=True)
return [early_stopping, tensorboard, checkpointer, csv_logger]
def main(mode):
if mode == 'hdf5':
print("Training from hdf5")
training_file = '../data/hdf5/training.h5'
validation_file = '../data/hdf5/validation.h5'
model_path = '../models/hdf5'
log_dir = '../logs/hdf5{0}'.format(strftime('%H%M%S'))
x_train, y_train, x_test, y_test = load_data_from_hdf5(training_file, validation_file)
elif mode == 'tfrecords':
print("Training from TFRecords")
training_file = '../data/tfrecords/training.tfrecords'
validation_file = '../data/tfrecords/validation.tfrecords'
model_path = '../models/tfrecords'
log_dir = '../logs/tfrecors{0}'.format(strftime('%H%M%S'))
x_train, y_train, x_test, y_test = load_data_from_tfrecords(training_file, validation_file)
else:
print("Pitty ...")
return
tensorboard = TensorBoard(log_dir=log_dir)
input_name = "image"
model = model_fn(IMAGE_SHAPE, INPUT_NAME)
x_train = x_train/255
x_test = x_test/255
y_train = tf.keras.utils.to_categorical(y_train, NUM_CLASSES)
y_test = tf.keras.utils.to_categorical(y_test, NUM_CLASSES)
model.fit(x_train, y_train, batch_size=32, epochs=5, verbose=1, callbacks=[tensorboard])
model_file = join(model_path, 'cnn_{0}'.format(strftime('%H%M%S')))
model.save(model_file)
results = model.evaluate(x_test, y_test)
print(results)
def train(self):
tbCallBack = TensorBoard(log_dir='./td_lstm_logs', histogram_freq=0, write_graph=True, write_images=True)
texts_raw_indices, texts_left_indices, aspects_indices, texts_right_indices, polarities_matrix = \
read_dataset(type=self.DATASET,
mode='test',
embedding_dim=self.EMBEDDING_DIM,
max_seq_len=self.MAX_SEQUENCE_LENGTH, max_aspect_len=self.MAX_ASPECT_LENGTH)
left_input = np.concatenate((texts_left_indices, aspects_indices), axis=1)
right_input = np.concatenate((texts_right_indices, aspects_indices), axis=1)
for i in range(1, self.ITERATION):
print()
print('-' * 50)
print('Iteration', i)
self.model.fit([self.left_input, self.right_input], self.polarities_matrix,
validation_data=([left_input, right_input], polarities_matrix),
batch_size=self.BATCH_SIZE, callbacks=[tbCallBack])
if i % 5 == 0:
self.model.save('td_lstm_saved_model.h5')
print('model saved')
def train_model(data, args):
train_config = TrainConfig(args)
mode_module = importlib.import_module("modes." + args.mode)
train_generator = mode_module.DataGenerator(data.train_data)
val_generator = mode_module.DataGenerator(data.validation_data)
model = mode_module.build_model()
model.compile(optimizer=train_config.optimizer,
loss=train_config.loss,
metrics=train_config.metrics)
results_csv_file = os.path.join(args.results_dir, "results.csv")
ckpt_filename = "Epoch-{epoch:02d}-Val-Acc-{val_accuracy:.4f}.hdf5"
weight_file = os.path.join(args.checkpoints_dir, ckpt_filename)
results_callback = CSVLogger(results_csv_file, append=True, separator=',')
checkpoints_callback = ModelCheckpoint(weight_file,
save_best_only=True,
save_weights_only=True)
tensorboard_callback = TensorBoard(log_dir=args.results_dir,
histogram_freq=0,
write_graph=True,
write_images=True)
model.fit_generator(generator=train_generator,
validation_data=val_generator,
verbose=2,
epochs=train_config.epochs,
shuffle=True,
callbacks=[
results_callback, tensorboard_callback,
checkpoints_callback
])
return model
def setup_callbacks(weights_log_path: str, batch_size: int,
lr: float) -> List[tf.keras.callbacks.Callback]:
"""
Sets up the callbacks for the training of the model.
"""
# Setup callback to save the best weights after each epoch
checkpointer = ModelCheckpoint(filepath=os.path.join(
weights_log_path, 'weights.{epoch:02d}-{val_loss:.2f}.hdf5'),
verbose=0,
save_best_only=True,
save_weights_only=True,
monitor='val_loss',
mode='min')
tensorboard_log_dir = os.path.join(weights_log_path, 'tensorboard')
# Note that update_freq is set to batch_size * 10,
# because the epoch takes too long and batch size too short
tensorboard = TensorBoard(log_dir=tensorboard_log_dir,
write_graph=True,
histogram_freq=0,
write_grads=True,
write_images=False,
batch_size=batch_size,
update_freq=batch_size * 10)
def lrs(epoch):
if epoch > 10:
return lr / 10
elif epoch > 6:
return lr / 5
else:
return lr
lr_schedule = LearningRateScheduler(lrs, verbose=1)
return [tensorboard, checkpointer, lr_schedule]
def __init__(self):
print(tf.__version__)
fashion_mnist = keras.datasets.fashion_mnist
(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
print("train_images shape is: %s".format(train_images.shape))
print("length of data: %s".format(len(train_labels)))
tensorboard = TensorBoard(log_dir="logs/{}".format(time()))
train_images = train_images / 255.0 # normalize data
test_images = test_images / 255.0 # normalize data
plt.figure(figsize=(10,10))
for i in range(25):
plt.subplot(5,5,i+1)
plt.xticks([])
plt.yticks([])
plt.grid(False)
plt.imshow(train_images[i], cmap=plt.cm.binary)
plt.xlabel(class_names[train_labels[i]])
model = keras.Sequential([
keras.layers.Flatten(input_shape=(28,28)),
keras.layers.Dense(128, activation=tf.nn.relu),
keras.layers.Dense(10, activation=tf.nn.softmax)
])
model.compile(optimizer=keras.optimizers.Adagrad(lr=0.1, epsilon=None, decay=0.0),
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
model.fit(train_images, train_labels, epochs=5,
callbacks=[tensorboard])
test_loss, test_acc = model.evaluate(test_images, test_labels)
print('Точность после проверки: ', test_acc)
predictions = model.predict(test_images)
model.save("fashion_mnist_model.h5")
def setup_callables(self):
monitor = "val_dice_coef"
# Setup callback to save best weights after each epoch
checkpointer = ModelCheckpoint(filepath=os.path.join(
self.model_dir, 'weights.{epoch:02d}-{val_loss:.2f}.hdf5'),
verbose=0,
save_best_only=True,
save_weights_only=True,
monitor=monitor,
mode='max')
# setup callback to register training history
csv_logger = CSVLogger(os.path.join(self.log_dir, 'log.csv'),
append=True,
separator=';')
# setup logger to catch warnings and info messages
set_logger(os.path.join(self.log_dir, 'train_val.log'))
# setup callback to retrieve tensorboard info
tensorboard = TensorBoard(log_dir=self.log_dir,
write_graph=True,
histogram_freq=0)
# setup early stopping to stop training if val_loss is not increasing after 3 epochs
early_stopping = EarlyStopping(monitor=monitor,
patience=5,
mode='max',
verbose=0)
lr_reducer = ReduceLROnPlateau(monitor=monitor,
factor=0.05,
cooldown=0,
patience=5,
verbose=0,
mode='max')
return [
checkpointer, csv_logger, tensorboard, early_stopping, lr_reducer
]
def get_densenet121_model(classes=2):
def preprocess_input(img):
img[:, :, 0] = (img[:, :, 0] - 103.94) * 0.017
img[:, :, 1] = (img[:, :, 1] - 116.78) * 0.017
img[:, :, 2] = (img[:, :, 2] - 123.68) * 0.017
return img.astype(np.float32)
def decode_img(img):
img[:, :, 0] = (img[:, :, 0] / 0.017) + 103.94
img[:, :, 1] = (img[:, :, 1] / 0.017) + 116.78
img[:, :, 2] = (img[:, :, 2] / 0.017) + 123.68
return img.astype(np.uint8)
base_model = tf.keras.applications.DenseNet121(include_top=False,
classes=2)
x = base_model.output
x = GlobalAveragePooling2D()(x)
pre = Dense(classes, activation='softmax', name='fc1000')(x)
model = Model(inputs=base_model.input, outputs=pre)
model.summary()
for layer in base_model.layers:
layer.trainable = False
ckpt = './ckpt/densenet121.h5'
checkpoint = ModelCheckpoint(filepath=ckpt)
tensorboard = './log/densenet121'
tensorboard = TensorBoard(log_dir=tensorboard)
if os.path.exists(ckpt):
model.load_weights(ckpt, by_name=True)
print("load done")
else:
plot_model(model, to_file='densenet121.png')
model.compile(optimizer=tf.train.AdamOptimizer(0.001),
loss='binary_crossentropy',
metrics=['accuracy'])
return model, checkpoint, tensorboard, preprocess_input, decode_img
def train_model(self,
model,
train_images,
train_labels,
val_images,
val_labels):
"""
"""
es = EarlyStopping(monitor='loss',
mode='max',
verbose=1,
patience=cfg.patience)
# modelcheckpoint
mc = ModelCheckpoint(super().get_path_to_save(),
monitor='acc',
mode='max',
save_best_only=True,
verbose=1)
# tensorborad
logdir = os.path.join(cfg.pfad_zu_logs, cfg.keras_model_name)
tb = TensorBoard(log_dir=logdir,
histogram_freq=0,
write_graph=True,
write_images=False)
callbak = [es, mc, tb]
self.compile_model(model)
print("______________Anfang des Trainings____________________")
history = model.fit(train_images,
train_labels,
validation_data=(val_images, val_labels),
epochs=self.__Num_epochs,
batch_size=self.__batch_size,
verbose=1,
validation_split=self.__validation_split,
callbacks=callbak)
print("training fertig")
return history
def fitModel(self, training_features: np.ndarray,
training_results_labels: np.ndarray, epochs: int) -> None:
"""
Takes the training set and its corresponding labels of the match outcome.
Uses Keras to perform propagation and backpropagation to train the network.
The final hyperparameters are set - batch size, epoch and callback to TensorBoard
for further visualisation.
:param training_features: NumPy array of the feature vectors (shape [10,])
:param training_results_labels: NumPy array of all the corresponding outcomes
:param epochs: Integer of the epoch count training will be carried through
"""
tensorboard = TensorBoard(log_dir="logs\{}".format(time()))
try:
self._model.fit(training_features,
training_results_labels,
epochs=epochs,
callbacks=[tensorboard],
batch_size=32)
except ValueError as e:
print(
"FAILED TRAINING, bad vector shape? not using NumPy array? :",
e)
exit(1)
def train_model(self, model, X_train, y_train):
start_time = time.time()
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
tensorboard = TensorBoard(log_dir=settings.lOGS_DIR)
early_stopping = keras.callbacks.EarlyStopping(monitor='val_loss',
min_delta=0.005,
patience=2,
verbose=0,
mode='auto')
history = model.fit(X_train,
y_train,
validation_split=0.2,
epochs=20,
callbacks=[tensorboard, early_stopping])
end_time = time.time()
print('Total train time = ', round(end_time - start_time), 's')
self._visualize_model_training(history)
return model
def makemod(LSTM_layers, LSTM_sizes, Dense_layers, loss_type="categorical_crossentropy", opt="adam"):
"""Defines, compiles and fits models"""
for lstmlayer in LSTM_layers:
for lstmsize in LSTM_sizes:
for denselayer in Dense_layers:
NAME = "{}-24 {}-LSTM-{}-Nodes-{}-Dense".format(text_designation, lstmlayer, lstmsize, denselayer)
model = Sequential()
for l in range(lstmlayer - 1):
model.add(LSTM(lstmsize, return_sequences=True, input_shape=(x_train.shape[1], x_train.shape[2])))
model.add(LSTM(lstmsize, input_shape=(x_train.shape[1], x_train.shape[2])))
for l in range(denselayer):
model.add(Dense(vocab_size, activation='relu'))
model.add(Dense(vocab_size, activation='softmax'))
print(model.summary())
# Log the model
tb = TensorBoard(log_dir="logs/{}".format(NAME))
# Compile model
model.compile(loss=loss_type, optimizer=opt, metrics=["accuracy"])
model.fit(x_train, y_train, epochs=1000, validation_split=0.1, verbose=2, callbacks=[tb])
print("Model {} created".format(NAME))
# Save Model
model.save(NAME)
print("Model {} saved".format(NAME))
def __train__(self, data_vector: {}, epochs: int = 15, log_dir=None):
"""
Trains a CNN on a given data vector
:param data_vector: training data vector
:param epochs: maximal number of epochs to train
:param log_dir: directory to save the log files, usually userdir/pickX/logs
:return: self
"""
callbacks = []
if log_dir is not None:
callbacks.append(TensorBoard(log_dir=log_dir))
callbacks.append(EarlyStopping("val_loss", patience=2))
self.model.fit(x=data_vector[self.feature_field],
y=data_vector['labels'],
validation_split=0.1,
batch_size=32,
epochs=epochs,
callbacks=callbacks)
self.model.summary()
self.input_dimension = self.model.input_shape[1:]
return self
def init_callbacks(self):
self.callbacks.append(
ModelCheckpoint(
filepath=os.path.join(self.config.callbacks.checkpoint_dir,
'%s-{epoch:02d}-{val_loss:.2f}.hdf5' % self.config.exp.name),
monitor=self.config.callbacks.checkpoint_monitor,
mode=self.config.callbacks.checkpoint_mode,
save_best_only=self.config.callbacks.checkpoint_save_best_only,
save_weights_only=self.config.callbacks.checkpoint_save_weights_only,
verbose=self.config.callbacks.checkpoint_verbose,
)
)
self.callbacks.append(
TensorBoard(
log_dir=self.config.callbacks.tensorboard_log_dir,
write_graph=self.config.callbacks.tensorboard_write_graph,
)
)
self.callbacks.append(
CSVLogger(os.path.join(self.config.callbacks.history_dir, "parameters.csv"), separator=',', append=False)
)
def forward(self, X_train, X_test, y_train, y_test):
X_shape = X_train.shape[1]
y_shape = y_train.shape[1]
X = Input(shape=(self.image_size, self.image_size, 1), name='input')
label = Input(shape=(y_shape,), name='label')
encoder, shape = self.encode(X, label)
encoder.summary()
z_inputs = Input(shape=(self.n_dim,), name='latent_input')
decoder = self.decode(z_inputs, label, shape)
decoder.summary()
z_output = encoder([X, label])[2]
outputs = decoder([z_output, label])
cvae = Model([X, label], outputs, name='cvae')
cvae.compile(optimizer=Adam(lr=self.learning_rate, decay=self.decay_rate, epsilon=1e-08), loss=self.vae_loss)
cvae.summary()
tensorboard = TensorBoard(log_dir="{}/{}".format(self.logs_dir,time()))
cvae_hist = cvae.fit([X_train, y_train], X_train, verbose=1, batch_size=self.batch_size, epochs=self.epochs,
validation_data=([X_test, y_test], X_test), callbacks=[tensorboard], shuffle=True)
decoder.save(self.args.save_model + '.h5')
return cvae, cvae_hist
def train(args):
'''
Train model
'''
model = build_model()
X_train, Y1_train, Y2_train, X_dev, Y1_dev, Y2_dev, X_test, Y1_test, Y2_test = load_dataset(
args.dataset_name)
mc = ModelCheckpoint('best_model.h5',
monitor='dense_2_categorical_accuracy',
mode='max',
verbose=1,
save_best_only=True)
es = EarlyStopping(monitor='loss', mode='min', verbose=1, patience=10)
print('Fitting model...')
results = model.fit(
X_train, [Y1_train, Y2_train],
epochs=args.epochs,
verbose=1,
validation_data=(X_dev, [Y1_dev, Y2_dev]),
callbacks=[TensorBoard(log_dir=TENSORBOARD_DIR), es, mc])
_, _, _, cat_acc, subcat_acc = model.evaluate(X_test, [Y1_test, Y2_test],
verbose=0)
print('last model:')
print(cat_acc)
print(subcat_acc)
model.save("model.h5")
print("Saved model to disk")
model1 = load_model('best_model.h5')
_, _, _, cat_acc, subcat_acc = model1.evaluate(X_test, [Y1_test, Y2_test],
verbose=0)
print('best model:')
print(cat_acc)
print(subcat_acc)
def get_callbacks(model_name: str) -> List[Union[TensorBoard, EarlyStopping, ModelCheckpoint]]:
"""Accepts the model name as a string and returns multiple callbacks for training the keras model.
Parameters
----------
model_name : str
The name of the model as a string.
Returns
-------
List[Union[TensorBoard, EarlyStopping, ModelCheckpoint]]
A list of multiple keras callbacks.
"""
logdir = (
"logs/scalars/" + model_name + "_" + datetime.now().strftime("%Y%m%d-%H%M%S")
) # create a folder for each model.
tensorboard_callback = TensorBoard(log_dir=logdir)
# use tensorboard --logdir logs/scalars in your command line to startup tensorboard with the correct logs
early_stopping_callback = EarlyStopping(
monitor="val_mean_absolute_percentage_error",
min_delta=1, # model should improve by at least 1%
patience=10, # amount of epochs with improvements worse than 1% until the model stops
verbose=2,
mode="min",
restore_best_weights=True, # restore the best model with the lowest validation error
)
model_checkpoint_callback = ModelCheckpoint(
"./data/models/" + model_name + ".h5",
monitor="val_mean_absolute_percentage_error",
verbose=0,
save_best_only=True, # save the best model
mode="min",
save_freq="epoch", # save every epoch
) # saving eff_net takes quite a bit of time
return [tensorboard_callback, early_stopping_callback, model_checkpoint_callback]
def train(self, train_gen, val_gen,
saved_model_path, epochs=100, steps=100, train_split=0.8,
verbose=1, min_delta=.0005, patience=5, use_early_stop=True):
"""
train_gen: generator that yields an array of images an array of
"""
# checkpoint to save model after each epoch
save_best = ModelCheckpoint(saved_model_path,
monitor='val_loss',
verbose=verbose,
save_best_only=True,
mode='min')
# stop training if the validation error stops improving.
early_stop = EarlyStopping(monitor='val_loss',
min_delta=min_delta,
patience=patience,
verbose=verbose,
mode='auto')
# callbacks_list = [save_best]
callbacks_list = [save_best, TensorBoard(log_dir='./tmp/log')]
if use_early_stop:
callbacks_list.append(early_stop)
hist = self.model.fit_generator(
train_gen,
steps_per_epoch=steps,
epochs=epochs,
verbose=1,
validation_data=val_gen,
callbacks=callbacks_list,
validation_steps=steps * (1.0 - train_split) / train_split)
return hist
