def get_optimizer(op_type, learning_rate):
if op_type == 'sgd':
return optimizers.SGD(learning_rate)
elif op_type == 'rmsprop':
return optimizers.RMSprop(learning_rate)
elif op_type == 'adagrad':
return optimizers.Adagrad(learning_rate)
elif op_type == 'adadelta':
return optimizers.Adadelta(learning_rate)
elif op_type == 'adam':
return optimizers.Adam(learning_rate, clipnorm=5)
elif op_type == 'adamw':
return AdamWeightDecay(
learning_rate=learning_rate,
weight_decay_rate=0.01,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
exclude_from_weight_decay=["layer_norm", "bias"])
elif op_type == 'adamw_2':
return create_optimizer(init_lr=learning_rate,
num_train_steps=9000,
num_warmup_steps=0)
elif op_type == 'adamw_3':
return create_optimizer(init_lr=learning_rate,
num_train_steps=9000,
num_warmup_steps=100)
else:
raise ValueError('Optimizer Not Understood: {}'.format(op_type))
def __init__(self,
path: str,
epochs: int = 50,
model: str or tf.keras.models.Sequential = None,
optimizer: tko.Optimizer = tko.Adadelta(lr=0.1),
threshold: float = None,
verbose: bool = False):
"""
Initialize audio downloader
:param path: working path
:param epochs: max training epochs
:param model: path to existing or keras Sequential
:param optimizer: keras optimizer
:param threshold: stop training if threshold exceeds given value
:param verbose: verbose output
"""
self.path = path
self.model = model
self.optimizer = optimizer
self.verbose = verbose
self.data = {}
self.model = model
self.threshold = threshold
self.epochs = epochs
def train():
global model, x_train, y_train
model.add(Conv2D(32,
kernel_size=(3, 3),
activation='relu',
padding="same",
input_shape=(imgW, imgH, 1)))
model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(classes_num, activation='softmax'))
model.summary()
print()
model.compile(loss=categorical_crossentropy,
optimizer=optimizers.Adadelta(),
metrics=['accuracy'])
model.fit(x=x_train,
y=y_train,
batch_size=batch_size,
verbose=2,
epochs=epochs,
validation_split=0.2)
model.save(filepath=modelfile)
def descent_optimizers():
optimizers.Adadelta(learning_rate=1e-3,
rho=0.95,
epsilon=1e-07,
name='Adadelta')
optimizers.Adagrad(learning_rate=1e-3,
initial_accumulator_value=0.1,
epsilon=1e-07,
name='Adagrad')
optimizers.Adam(learning_rate=1e-3,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-07,
amsgrad=False,
name='Adam')
optimizers.Adamax(learning_rate=1e-3,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-07,
name='Adamax')
optimizers.Nadam(learning_rate=1e-3,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-07,
name='Nadam')
optimizers.RMSprop(learning_rate=1e-3,
rho=0.9,
momentum=0.0,
epsilon=1e-07,
centered=False,
name='RMSprop')
optimizers.SGD(learning_rate=1e-2,
momentum=0.0,
nesterov=False,
name='SGD')
def main():
#file = r'./db/fucDatasetReg_1F_NoLinear.csv'
#file = r'./db/fucDatasetReg_2F.csv'
file = r'./db/fucDatasetReg_3F_1000.csv'
x_train, x_test, y_train, y_test = getCsvDataset(file)
lr = 1e-3
EPOCHES = 200
# optimizer = optimizerTf(lr=lr)
# losses,_ = trainModel(x_train,y_train,optimizer,epochs=EPOCHES)
# plotLoss(losses)
opts = []
# fast group
opts.append((optimizers.SGD(learning_rate=lr), 'SGD'))
opts.append((optimizers.RMSprop(learning_rate=lr), 'RMSprop'))
opts.append((optimizers.Adam(learning_rate=lr), 'Adam'))
opts.append((optimizers.Adamax(learning_rate=lr), 'Adamax'))
opts.append((optimizers.Nadam(learning_rate=lr), 'Nadam'))
# # slow group
opts.append((optimizers.Adadelta(learning_rate=lr), 'Adadelta'))
opts.append((optimizers.Adagrad(learning_rate=lr), 'Adagrad'))
opts.append((optimizers.Ftrl(learning_rate=lr), 'Ftrl'))
lossesDict = {}
for opti, name in opts:
losses, _ = trainModel(x_train, y_train, opti, epochs=EPOCHES)
lossesDict[name] = losses
#print(name, losses)
plotLossDict(lossesDict)
def build_optimizer(type, lr, kerasDefaults):
""" Set the optimizer to the appropriate Keras optimizer function
based on the input string and learning rate. Other required values
are set to the Keras default values
Parameters
----------
type : string
String to choose the optimizer
Options recognized: 'sgd', 'rmsprop', 'adagrad', adadelta', 'adam'
See the Keras documentation for a full description of the options
lr : float
Learning rate
kerasDefaults : list
List of default parameter values to ensure consistency between frameworks
Returns
----------
The appropriate Keras optimizer function
"""
if type == 'sgd':
return optimizers.SGD(lr=lr, decay=kerasDefaults['decay_lr'],
momentum=kerasDefaults['momentum_sgd'],
nesterov=kerasDefaults['nesterov_sgd']) # ,
# clipnorm=kerasDefaults['clipnorm'],
# clipvalue=kerasDefaults['clipvalue'])
elif type == 'rmsprop':
return optimizers.RMSprop(lr=lr, rho=kerasDefaults['rho'],
epsilon=kerasDefaults['epsilon'],
decay=kerasDefaults['decay_lr']) # ,
# clipnorm=kerasDefaults['clipnorm'],
# clipvalue=kerasDefaults['clipvalue'])
elif type == 'adagrad':
return optimizers.Adagrad(lr=lr,
epsilon=kerasDefaults['epsilon'],
decay=kerasDefaults['decay_lr']) # ,
# clipnorm=kerasDefaults['clipnorm'],
# clipvalue=kerasDefaults['clipvalue'])
elif type == 'adadelta':
return optimizers.Adadelta(lr=lr, rho=kerasDefaults['rho'],
epsilon=kerasDefaults['epsilon'],
decay=kerasDefaults['decay_lr']) # ,
# clipnorm=kerasDefaults['clipnorm'],
# clipvalue=kerasDefaults['clipvalue'])
elif type == 'adam':
return optimizers.Adam(lr=lr, beta_1=kerasDefaults['beta_1'],
beta_2=kerasDefaults['beta_2'],
epsilon=kerasDefaults['epsilon'],
decay=kerasDefaults['decay_lr']) # ,
def get_keras_optimizer(optimizer_params: Dict):
from tensorflow.keras import optimizers #pylint: disable=import-error
if optimizer_params['type'] == 'sgd':
return optimizers.SGD(**optimizer_params['kwargs'])
if optimizer_params['type'] == 'adadelta':
return optimizers.Adadelta(**optimizer_params['kwargs'])
if optimizer_params['type'] == 'adam':
return optimizers.Adam(**optimizer_params['kwargs'])
raise ValueError('The optimizer {} is not supported.'.format(
optimizer_params['type']))
def callOptimizer(opt='rmsprop'):
'''Function returns the optimizer to use in .fit()
options:
adam, sgd, rmsprop, ada_grad,ada_delta,ada_max
'''
opt_dict = {'adam': optimizers.Adam(),
'sgd' : optimizers.SGD(),
'rmsprop' : optimizers.RMSprop(),
'ada_grad' : optimizers.Adagrad(),
'ada_delta': optimizers.Adadelta(),
'ada_max' : optimizers.Adamax()}
return opt_dict[opt]
def definePacmanTestModel1(conf):
# Define Model
inputShape = (conf.input_y_dim, conf.input_x_dim, conf.c_channels)
state = Input(inputShape) # pre 0 in
x = Conv2D(32, (3, 3), activation='relu', padding='same',
name='Conv0')(state) # conv 0
x = Conv2D(8, (3, 3), activation='relu', padding='valid',
name='Conv1')(x) # conv 1
x = MaxPooling2D((2, 2))(x) # pooling
x = Conv2D(8, (3, 3), activation='relu', padding='valid',
name='Conv2')(x) # conv 2
x = MaxPooling2D((2, 2))(x) # pooling
x = Conv2D(8, (3, 3), activation='relu', padding='valid',
name='Conv3')(x) # conv 3
x = MaxPooling2D((2, 2))(x) # pooling
x = Flatten()(x) # flatten
x = Dense(100, activation='relu')(x) # fc
x = Dense(100, activation='relu')(x) # fc
qsa = Dense(conf.num_actions, activation='linear')(x) # out
# Make Model
model = Model(state, qsa)
# Configure Optimizer
if conf.optimizer == 'adadelta':
optimizer = optimizers.Adadelta(lr=conf.learning_rate,
decay=0.0,
rho=0.95)
elif conf.optimizer == 'sgd':
optimizer = optimizers.SGD(lr=conf.learning_rate)
elif conf.optimizer == 'adam':
optimizer = optimizers.Adam(lr=conf.learning_rate)
elif conf.optimizer == 'adagrad':
optimizer = optimizers.Adagrad(lr=conf.learning_rate)
else:
print("Optimizer '{0}' not found.".format(conf.optimizer))
exit(0)
return model, optimizer
def create_optimizer(opt,
learning_rate,
momentum=0.9,
decay=0.0,
nesterov=False):
"""
Create optimizer operation
:param opt: A string which can be one of 'sgd', 'momentum' or 'adam'
:param learning_rate: A float value
:param momentum: A float value
:return: An optimizer operation
"""
assert opt in [
'sgd', 'rmsprop', 'adagrad', 'adadelta', 'adam', 'adamax', 'nadam'
]
if opt == 'sgd':
optimizer = optimizers.SGD(lr=learning_rate,
momentum=momentum,
decay=decay,
nesterov=nesterov)
elif opt == 'rmsprop':
optimizer = optimizers.RMSprop(lr=learning_rate,
rho=0.9,
epsilon=1e-06)
elif opt == 'adagrad':
optimizer = optimizers.Adagrad(lr=learning_rate, epsilon=1e-06)
elif opt == 'adadelta':
optimizer = optimizers.Adadelta(lr=learning_rate,
rho=0.95,
epsilon=1e-06)
elif opt == 'adam':
optimizer = optimizers.Adam(lr=learning_rate,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08)
elif opt == 'adamax':
optimizer = optimizers.Adamax(lr=learning_rate,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08)
elif opt == 'nadam':
optimizer = optimizers.Nadam(lr=learning_rate,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08,
schedule_decay=0.004)
else:
optimizer = None
return optimizer
def deep_network(self, pca=None):
if not pca:
print("Regular")
model = Sequential()
model.add(Dense(64, activation='relu', input_dim=self.train_x_.shape[1]))
model.add(Dense(16, activation='relu'))
model.add(Dense(self.classes, activation='relu'))
opt = optimizers.Adadelta(learning_rate=1.0, rho=0.95)
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
print("samples: ", self.train_x_.shape, "classes: ", self.train_y_cat.shape)
model.fit(self.train_x_, self.train_y_cat, validation_split=0.2, batch_size=32, epochs=50)
model.evaluate(self.test_x_, self.test_y_cat)
# model_prediction = model.predict(self.test_x_, batch_size=32, verbose=1)
#
# return model_prediction
elif pca:
print("PCA")
model_pca = Sequential()
model_pca.add(Dense(64, activation='tanh', input_dim=self.train_x_pca_.shape[1]))
model_pca.add(Dropout(0.25))
model_pca.add(Dense(16, activation='tanh'))
model_pca.add(Dense(self.classes, activation='softmax'))
opt = optimizers.Adadelta(learning_rate=1.0, rho=0.95)
model_pca.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
model_pca.fit(self.train_x_pca_, self.train_y_cat, validation_split=0.2, batch_size=32, epochs=10)
model_pca.evaluate(self.test_x_pca_, self.test_y_cat)
def __get_optimizer(optimizer, lr):
if optimizer == 'sgd':
return optimizers.SGD(lr=lr)
elif optimizer == 'rmsprop':
return optimizers.RMSprop(lr=lr)
elif optimizer == 'adagrad':
return optimizers.Adagrad(lr=lr)
elif optimizer == 'adadelta':
return optimizers.Adadelta(lr=lr)
elif optimizer == 'adam':
return optimizers.Adam(lr=lr)
elif optimizer == 'adamax':
return optimizers.Adamax(lr=lr)
elif optimizer == 'nadam':
return optimizers.Nadam(lr=lr)
def _compile(self,
model,
loss_function,
optimizer,
lr=0.01,
decay=0.0,
clipnorm=0.0):
"""Compiles a model specified with Keras.
See https://keras.io/optimizers/ for more info on each optimizer.
Args:
model: Keras model object to compile
loss_function: Keras loss_function object to compile model with
optimizer (str): the optimizer to use during training
lr (float): learning rate to use during training
decay (float): per epoch decay rate
clipnorm (float): gradient normalization threshold
"""
# The parameters of these optimizers can be freely tuned.
if optimizer == 'sgd':
optimizer_ = optimizers.SGD(lr=lr, decay=decay, clipnorm=clipnorm)
elif optimizer == 'adam':
optimizer_ = optimizers.Adam(lr=lr, decay=decay, clipnorm=clipnorm)
elif optimizer == 'adamax':
optimizer_ = optimizers.Adamax(lr=lr,
decay=decay,
clipnorm=clipnorm)
# It is recommended to leave the parameters of this optimizer at their
# default values (except the learning rate, which can be freely tuned).
# This optimizer is usually a good choice for recurrent neural networks
elif optimizer == 'rmsprop':
optimizer_ = optimizers.RMSprop(lr=lr, clipnorm=clipnorm)
# It is recommended to leave the parameters of these optimizers at their
# default values.
elif optimizer == 'adagrad':
optimizer_ = optimizers.Adagrad(clipnorm=clipnorm)
elif optimizer == 'adadelta':
optimizer_ = optimizers.Adadelta(clipnorm=clipnorm)
elif optimizer == 'nadam':
optimizer_ = optimizers.Nadam(clipnorm=clipnorm)
else:
err_msg = "Argument for `optimizer` is invalid, got: {}".format(
optimizer)
LOGGER.error('ValueError %s', err_msg)
raise ValueError(err_msg)
model.compile(optimizer=optimizer_, loss=loss_function)
def construct_model():
input1 = Input(shape=(seq_length, 5790, 6))
input2 = Input(shape=(seq_length, 6))
re_input1 = Reshape((5790 * seq_length, 6))(input1)
rere_input1 = Permute((2, 1), input_shape=(5790 * seq_length, 6))(re_input1)
conv1 = Conv1D(30, 1, strides=1, padding='valid', activation='relu', data_format="channels_first", name='X1_input')(
rere_input1)
conv2 = Conv1D(30, 1, strides=1, padding='valid', activation='relu', data_format="channels_first", name='Conv7')(
conv1)
LSTM1 = LSTM(4, return_sequences=True)(input2)
LSTM2 = LSTM(4, return_sequences=False)(LSTM1)
reshape_conv2 = Reshape((30, 5790, seq_length))(conv2)
pool = MaxPooling2D(pool_size=(1, 2), strides=(1, 2), padding='valid', data_format="channels_last")(reshape_conv2)
reshape1 = Reshape((1, 30, 2895, seq_length))(pool)
reshape2 = Permute((4, 2, 3, 1), input_shape=(1, 30, 2895, seq_length))(reshape1)
convLSTM1 = ConvLSTM2D(filters=10, kernel_size=(3, 3), strides=(3, 3),
padding='same', return_sequences=True)(reshape2)
convLSTM2 = ConvLSTM2D(filters=20, kernel_size=(3, 2), strides=(2, 2),
padding='same', return_sequences=True)(convLSTM1)
convLSTM3 = ConvLSTM2D(filters=40, kernel_size=(3, 1), strides=(2, 2),
padding='same', return_sequences=True)(convLSTM2)
convLSTM4 = ConvLSTM2D(filters=40, kernel_size=(2, 2), strides=(2, 2),
padding='same', return_sequences=False)(convLSTM3)
flat1 = Flatten()(convLSTM4)
flat2 = Flatten()(LSTM2)
dense1 = Dense(120)(flat1)
activation1 = Activation('relu')(dense1)
merge2 = concatenate([activation1, flat2])
dense2 = Dense(30)(merge2)
activation2 = Activation('relu')(dense2)
output = Dense(1, kernel_regularizer=regularizers.l2(0.000001))(activation2)
model = Model(inputs=[input1, input2], outputs=[output])
sgd = optimizers.Adadelta(lr=1.0, rho=0.95, epsilon=None, decay=0.0)
model = multi_gpu_model(model, gpus=2)
model.compile(loss='mean_squared_error', optimizer=sgd)
print(model.summary())
return model
def get_optimizer(args):
clipvalue = 0
clipnorm = 10
if args.algorithm == 'rmsprop':
optimizer = opt.RMSprop(lr=0.001,
rho=0.9,
epsilon=1e-06,
clipnorm=clipnorm,
clipvalue=clipvalue)
elif args.algorithm == 'sgd':
optimizer = opt.SGD(lr=0.01,
momentum=0.0,
decay=0.0,
nesterov=False,
clipnorm=clipnorm,
clipvalue=clipvalue)
elif args.algorithm == 'adagrad':
optimizer = opt.Adagrad(lr=0.01,
epsilon=1e-06,
clipnorm=clipnorm,
clipvalue=clipvalue)
elif args.algorithm == 'adadelta':
optimizer = opt.Adadelta(lr=1.0,
rho=0.95,
epsilon=1e-06,
clipnorm=clipnorm,
clipvalue=clipvalue)
elif args.algorithm == 'adam':
optimizer = opt.Adam(lr=0.001,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08,
clipnorm=clipnorm,
clipvalue=clipvalue)
elif args.algorithm == 'adamax':
optimizer = opt.Adamax(lr=0.002,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08,
clipnorm=clipnorm,
clipvalue=clipvalue)
else:
raise Exception("Can't find optimizer " + args.algorithm)
return optimizer
def get_optimizer(self):
"""
Returns tf.keras.optimizer based on config
:return: optimizer
"""
if self.optimizer == 'Adam':
return opt.Adam(learning_rate=self.learning_rate)
elif self.optimizer == 'Adadelta':
return opt.Adadelta(learning_rate=self.learning_rate)
elif self.optimizer == 'RMSprop':
return opt.RMSprop(learning_rate=self.learning_rate)
elif self.optimizer == 'SGD':
return opt.SGD(learning_rate=self.learning_rate,
momentum=self.config.sgd_momentum)
else:
raise ValueError("%s optimizer not found in tf.keras.optimizers" %
self.optimizer)
def autoencoder_stacked(entrada,
train_set,
test_set,
dim_latente=250,
epochs=125,
lr=0.30,
grabar=False,
cargar=False,
archivo_encoder=" ",
archivo_autoencoder=" ",
grafica=False):
if cargar:
encoder = load_model(archivo_encoder)
autoencoder_deep = load_model(archivo_autoencoder)
else:
dim_input = entrada.shape[1]
adadelta = optimizers.Adadelta(lr=lr, rho=0.95)
input = Input(shape=((dim_input, )))
encoded = Dense(2000, activation='selu')(input)
encoded = Dense(1000, activation='selu')(encoded)
encoded = Dense(dim_latente, activation='selu')(encoded)
decoded = Dense(1000, activation='selu')(encoded)
decoded = Dense(2000, activation='selu')(decoded)
decoded = Dense(dim_input, activation='sigmoid')(decoded)
autoencoder_deep = Model(input, decoded)
encoder = Model(input, encoded)
autoencoder_deep.compile(optimizer=adadelta,
loss='binary_crossentropy',
metrics=["accuracy"])
autoencoder_deep.fit(train_set,
train_set,
epochs=epochs,
batch_size=256,
shuffle=True,
validation_data=(test_set, test_set))
if grafica:
galassify_grafica_error.grafica_loss(autoencoder_deep, epochs)
if grabar:
autoencoder_deep.save(archivo_autoencoder)
encoder.save(archivo_encoder)
return encoder, autoencoder_deep
def set_optimizers(self, select_optimizer, select_lr):
print("optimizers setting.")
print("optimizers : ", select_optimizer)
# Configure the optimizer from config.
if select_optimizer == "adam" or select_optimizer == "Adam":
opt = optimizers.Adam(lr=select_lr)
elif select_optimizer == "sgd" or select_optimizer == "SGD":
opt = optimizers.SGD(lr=select_lr)
elif select_optimizer == "adagrad" or select_optimizer == "Adagrad":
opt = optimizers.Adagrad(lr=select_lr)
elif select_optimizer == "adadelta" or select_optimizer == "Adadelta":
opt = optimizers.Adadelta(lr=select_lr)
else:
print("This is an unconfigured optimizer that uses Adam instead.")
opt = optimizers.Adam(lr=select_lr)
print("optimizer setting ... ok.")
return opt
def cc_optimizer(self, learning_rate, decay_rate=0, optimizer='adam'):
if optimizer == 'sgd':
self.cc_optimizer = optimizers.SGD(lr=learning_rate,\
decay = decay_rate, \
momentum = moment, \
nesterov=True)
elif optimizer == 'rms':
#--------------------------------------------------------------
self.cc_optimizer = optimizers.RMSprop(lr = learning_rate, \
rho= 0.9, \
epsilon = None,\
decay = decay_rate)
elif optimizer == 'adagrad':
#--------------------------------------------------------------
self.cc_optimizer = optimizers.Adagrad (lr = learning_rate , \
epsilon = None , \
decay = decay_rate)
elif optimizer == 'adadelta':
#--------------------------------------------------------------
self.cc_optimizer = optimizers.Adadelta(lr = learning_rate, \
rho=0.95 , \
epsilon = None,\
decay = decay_rate)
elif optimizer == 'nadam':
self.cc_optimizer = optimizers.Nadam(lr = learning_rate, \
beta_1 = 0.9, \
beta_2 = 0.999, \
epsilon = None, \
schedule_decay = 0.004)
else:
self.cc_optimizer = optimizers.Adam(lr = learning_rate, \
beta_1 = 0.9 , \
beta_2 = 0.999 , \
epsilon = None,\
decay = decay_rate,\
amsgrad = True )
return self.cc_optimizer
def set_optimizer(self, optimizer, loss="categorical_crossentropy"):
if optimizer.lower() == "adam":
opt_handle = Opt.Adam()
elif optimizer.lower() == "adagrad":
opt_handle = Opt.Adagrad()
elif optimizer.lower() == "adadelta":
opt_handle = Opt.Adadelta()
elif optimizer.lower() == "rmsprop":
opt_handle = Opt.RMSprop()
else:
print("Unknown optimizer {}. Using Adam!".format(optimizer))
opt_handle = Opt.Adam()
print("Setting model optimizer to {}".format(optimizer))
self.model.compile(
loss="categorical_crossentropy",
optimizer=opt_handle,
metrics=["accuracy"],
)
def setOptimizer(self, config):
configOptimizer = config["model"]["optimizer"].lower()
if configOptimizer == "Adadelta".lower():
self.optimizer = optimizers.Adadelta()
elif configOptimizer == "Adagrad".lower():
self.optimizer = optimizers.Adagrad()
elif configOptimizer == "Adamax".lower():
self.optimizer = optimizers.Adamax()
elif configOptimizer == "Ftrl".lower():
self.optimizer = optimizers.Ftrl()
elif configOptimizer == "SGD".lower():
self.optimizer = optimizers.SGD()
elif configOptimizer == "Nadam".lower():
self.optimizer = optimizers.Nadam()
elif configOptimizer == "Optimizer".lower():
self.optimizer = optimizers.Optimizer()
elif configOptimizer == "RMSprop".lower():
self.optimizer = optimizers.RMSprop()
def create_model(jets_dim, train_var, FC_layers, lr, beta, lamb, seed, encoder, n_gpus):
if len(set(train_var)-{'constituents'}) == 0: input_dim = jets_dim
elif 'constituents' not in train_var : input_dim = len(train_var)
else : input_dim = jets_dim+len(train_var)-1
tf.debugging.set_log_device_placement(False)
strategy = tf.distribute.MirroredStrategy(devices=['/gpu:'+str(n) for n in np.arange(n_gpus)])
with strategy.scope():
#loss = 'mean_squared_error'
loss = 'binary_crossentropy'
if encoder == 'dual_ae':
model = dual_AE(jets_dim, scalars_dim, FC_layers)
model.compile(optimizer=optimizers.Adadelta(lr=lr), loss=loss, loss_weights=[1.0,1.0])
else:
if encoder == 'oe_vae': model = FC_VAE(input_dim, FC_layers, beta, lamb, seed)
if encoder == 'vae' : model = FC_VAE(input_dim, FC_layers, beta, seed)
if encoder == 'ae' : model = FC_AE (input_dim, FC_layers)
model.compile(optimizer=optimizers.Adam(lr=lr, amsgrad=False), loss=loss)
#model.compile(optimizer=optimizers.Adam(lr=lr, amsgrad=False), loss=loss, loss_weights=[1.0,1.0])
print('\nNEURAL NETWORK ARCHITECTURE'); model.summary(); print()
return model
def get_optimizer():
optimizer_name = optimizer_names[random.randint(0, len(optimizer_names) - 1)]
model_attributes.optimizer_name = optimizer_name
if optimizer_name == 'SGD':
return optimizers.SGD(lr=get_learning_rate())
elif optimizer_name == 'RMSprop':
return optimizers.RMSprop(lr=get_learning_rate())
elif optimizer_name == 'Adagrad':
return optimizers.Adagrad(lr=get_learning_rate())
elif optimizer_name == 'Adadelta':
return optimizers.Adadelta(lr=get_learning_rate())
elif optimizer_name == 'Adam':
return optimizers.Adam(lr=get_learning_rate())
elif optimizer_name == 'Adamax':
return optimizers.Adamax(lr=get_learning_rate())
elif optimizer_name == 'Nadam':
return optimizers.Nadam(lr=get_learning_rate())
return None
def create_optimizer(optimizer, learning_rate):
"""
Simply returns an optimizer based on the string refering it.
:param optimizer: Name of the optimizer
:param learning_rate: learning rate of the optimizer
:return: tensorflow.keras.{optimizers}.
"""
if optimizer.lower() == "adadelta":
return optimizers.Adadelta(lr=learning_rate)
elif optimizer.lower() == "adagrad":
return optimizers.Adagrad(lr=learning_rate)
elif optimizer.lower() == "adam":
return optimizers.Adam(lr=learning_rate)
elif optimizer.lower() == "adamax":
return optimizers.Adamax(lr=learning_rate)
elif optimizer.lower() == "nadam":
return optimizers.Nadam(lr=learning_rate)
elif optimizer.lower() == "rmsprop":
return optimizers.RMSprop(lr=learning_rate)
elif optimizer.lower() == "sgd":
return optimizers.SGD(lr=learning_rate)
def select_optimizer(optimizer: str, learning_rate: float,
clipnorm: float) -> object:
"""Sets the optimizer for the training process of the neural network using the specified parameters.
Input
:param optimizer: name of the optimizer algorithm to be used (options: ADAM, SGD, RMSPROP, ADADELTA, Nesterov-ADAM)
:type optimizer: str
:param learning_rate: initial learning rate of the optimzier for neural network training
:type learning_rate: float
:param clipnorm: gradients will be clipped when their L2 norm exceeds this value.
:type clipnorm: float
Output
:return: abstract optimizer base class of the keras package
:rtype: object
"""
if optimizer == 'ADAM':
optimizer = optimizers.Adam(
lr=learning_rate, clipnorm=clipnorm) # , beta_1=0.9, beta_2=0.999)
if optimizer == 'SGD':
optimizer = optimizers.SGD(lr=learning_rate,
nesterov=True,
clipnorm=clipnorm)
if optimizer == 'RMSPROP':
optimizer = optimizers.RMSprop(lr=learning_rate,
momentum=0.9,
clipnorm=clipnorm)
if optimizer == 'ADADELTA':
optimizer = optimizers.Adadelta(lr=learning_rate) # , rho=0.95)
if optimizer == 'Nesterov-ADAM':
optimizer = optimizers.Nadam(
lr=learning_rate,
clipnorm=clipnorm) # , beta_1=0.91, beta_2=0.997)
return optimizer
def chexnet_model(FLAGS):
""" Builds the chexnet model using specifics from FLAGS. Returns a compiled model."""
base_model = DenseNet121(include_top=False,
weights='imagenet',
input_shape=(FLAGS.image_size, FLAGS.image_size,
3))
x = base_model.output
x = GlobalAveragePooling2D()(x)
predictions = Dense(14, activation='sigmoid', bias_initializer='ones')(x)
model = Model(inputs=base_model.input, outputs=predictions)
if FLAGS.opt == 'adam':
opt = optimizers.Adam(lr=FLAGS.lr)
elif FLAGS.opt == 'sgd':
opt = optimizers.SGD(lr=FLAGS.lr,
momentum=FLAGS.momentum,
nesterov=FLAGS.nesterov)
elif FLAGS.opt == 'rmsprop':
opt = optimizers.RMSProp(lr=FLAGS.lr)
elif FLAGS.opt == 'adagrad':
opt = optimizers.Adagrad(lr=FLAGS.lr)
elif FLAGS.opt == 'adadelta':
opt = optimizers.Adadelta(lr=FLAGS.lr)
elif FLAGS.opt == 'adamax':
opt = optimizers.Adamax(lr=FLAGS.lr)
elif FLAGS.opt == 'nadam':
opt = optimizers.Nadam(lr=FLAGS.lr)
else:
print("No optimizer selected. Using Adam.")
opt = optimizers.Adam(lr=FLAGS.lr)
hvd_opt = hvd.DistributedOptimizer(opt)
model.compile(loss='binary_crossentropy',
optimizer=hvd_opt,
metrics=['accuracy'])
return model
def get_optimizer(optimization_function, learning_rate):
if optimization_function == "Adam":
optimization_function = Optimizer.Adam(learning_rate=learning_rate)
elif optimization_function == "SGD":
optimization_function = Optimizer.SGD(learning_rate=learning_rate)
elif optimization_function == "RMSprop":
optimization_function = Optimizer.RMSprop(learning_rate=learning_rate)
elif optimization_function == "Adagrad":
optimization_function = Optimizer.Adagrad(learning_rate=learning_rate)
elif optimization_function == "Adadelta":
optimization_function = Optimizer.Adadelta(learning_rate=learning_rate)
elif optimization_function == "Adamax":
optimization_function = Optimizer.Adamax(learning_rate=learning_rate)
elif optimization_function == "Nadam":
optimization_function = Optimizer.Nadam(learning_rate=learning_rate)
return optimization_function
print(y_train.shape)
print(x_train.shape)
model = Sequential() #更正：不是model.Sequential()
input_shape = [im_row, im_col, 1]
model.add(
Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=input_shape))
model.add(Conv2D(64, kernel_size=(3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2))) #更正：这里是pool_size 不是kernel_size
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.25))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss=losses.categorical_crossentropy,
optimizer=optimizers.Adadelta(),
metrics=['accuracy']) #这里的optimizers 有s
history = model.fit(x_train,
y_train,
batch_size=batch_size,
epochs=epochs,
verbose=2,
validation_data=(x_test, y_test)) #注意是epochs 不是es
# plt.plot(history.history['loss'])
# plt.plot(history.history['val_loss'])
# plt.title('Model loss')
# plt.ylabel('loss')
# plt.xlabel('Epoch')
# plt.legend(['Train', 'Test'], loc='upper left')
# plt.show()
print(Utils.classificationReport('test', db.y_test, y_pred))
Utils.rocCurve(
db.name + '_' + schema.name + '_' + estimator.name + '_Augmented',
db.Y_test(), estimator.predict(db.X_test), db.info['n_cls'])
schema.saveWeights(estimator.name + '_' + db.name + '_Augmented')
schema.extract(estimator.name + '_Augmented', db)
for data, version, augment in Conf:
db = Loader.getDataset(data)
db = Utils.preProcessing(db)
# db.histogram()
db.summary()
# %% Conventional
# NotAugmentedRun(db, version, Estm.Conventional,
# losses.categorical_crossentropy,
# optimizers.Adadelta(), ['acc'],
# [callbacks.EarlyStopping(patience=20)])
generator = General(X_train=db.X_train,
y_train=db.y_train,
augment=True,
allowable=augment)
AugmentedRun(db, generator, version,
Estm.Conventional, losses.categorical_crossentropy,
optimizers.Adadelta(), ['acc'],
[callbacks.EarlyStopping(patience=20)])
def train_ae_coea(net_params, layer_params_list, data_train, data_eval, n_layers=4,
iters=4000, ind=None):
"""Trains the Autoencoder with the given parameters. Returns max Rho_MK and validation loss"""
K.clear_session()
# Required in order to have reproducible results from a specific random seed
os.environ['PYTHONHASHSEED'] = '0'
# Force tf to use a single thread (required for reproducibility)
session_conf = tf.compat.v1.ConfigProto(intra_op_parallelism_threads=1,
inter_op_parallelism_threads=1)
sess = tf.compat.v1.Session(graph=tf.compat.v1.get_default_graph(),
config=session_conf)
tf.compat.v1.keras.backend.set_session(sess)
# network parameters
batch = 32
optim = net_params['optim']
learn_rate = net_params['learn_rate']
decay = net_params['decay']
mom = net_params['mom']
rand_seed = net_params['rand_seed']
np.random.seed(rand_seed)
rn.seed(rand_seed)
tf.compat.v1.set_random_seed(rand_seed)
iter_per_epoch = int(np.ceil(len(data_train) / batch))
epochs = int(np.ceil(iters / iter_per_epoch))
if optim == 'adam':
opt = optimizers.Adam(lr=learn_rate, beta_1=mom, decay=decay, clipvalue=0.3)
elif optim == 'nadam':
opt = optimizers.Nadam(lr=learn_rate, beta_1=mom, schedule_decay=decay, clipvalue=0.3)
elif optim == 'rmsprop':
opt = optimizers.RMSprop(lr=learn_rate, rho=mom, decay=decay, clipvalue=0.3)
else: # adadelta
opt = optimizers.Adadelta(lr=learn_rate, rho=mom, decay=decay, clipvalue=0.3)
# pretraining
input_data = data_train
weights = []
ea = EarlyStopping(patience=int(epochs / 3))
cb = [ea]
for i, layer_params in enumerate(layer_params_list):
layer_params = layer_params_list[n_layers - 1 - i]
input_layer = layers.Input(shape=(input_data.shape[1],))
hidden_layer = layers.Dense(layer_params['n_neuron'],
activation=layer_params['act'],
kernel_regularizer=regularizers.l2(layer_params['L2']),
activity_regularizer=SparseActivityRegularizer
(p=layer_params['SP'], sparsity_beta=layer_params['SR']),
kernel_initializer=layer_params['init'])(input_layer)
output_layer = layers.Dense(input_data.shape[1], activation=layer_params['act'],
kernel_initializer=layer_params['init'])(hidden_layer)
model = models.Model(input_layer, output_layer)
model.compile(optimizer=opt, loss='mse')
history = model.fit(x=input_data, y=input_data, batch_size=batch,
epochs=epochs,
callbacks=cb, validation_split=0.2,
verbose=False)
for loss in history.history['loss']:
if np.isnan(loss):
K.clear_session()
return 0.01, 100
h_weights = model.get_weights()
weights.insert(2 * i, h_weights[0])
weights.insert(2 * i + 1, h_weights[1])
weights.insert(len(weights) - 2 * i, h_weights[2])
weights.insert(len(weights) - 2 * i, h_weights[3])
model = models.Model(input_layer, hidden_layer)
input_data = model.predict(input_data)
# stacking the layers - fine tuning
input_layer = layers.Input(shape=(data_train.shape[1],))
enc = layers.Dense(layer_params_list[-1]['n_neuron'],
activation=layer_params_list[-1]['act'],
kernel_initializer=layer_params_list[-1]['init'])(input_layer)
for i in range(n_layers - 1):
enc = layers.Dense(layer_params_list[-2 - i]['n_neuron'],
activation=layer_params_list[-2 - i]['act'],
kernel_initializer=layer_params_list[-2 - i]['init'])(enc)
dec = layers.Dense(layer_params_list[1]['n_neuron'],
activation=layer_params_list[1]['act'],
kernel_initializer=layer_params_list[1]['init'])(enc)
for i in range(n_layers - 2):
dec = layers.Dense(layer_params_list[i + 2]['n_neuron'],
activation=layer_params_list[i + 2]['act'],
kernel_initializer=layer_params_list[i + 2]['init'])(dec)
output_layer = layers.Dense(len(data_train.T),
activation=layer_params_list[-1]['act'],
kernel_initializer=layer_params_list[-1]['init'])(dec)
# assumption: output layer has the same parameters as the final hidden layer
model = models.Model(input_layer, output_layer)
model.compile(optimizer=opt, loss='mse')
model.set_weights(weights)
history = model.fit(x=data_train, y=data_train, batch_size=batch,
epochs=epochs,
callbacks=cb, validation_data=(data_eval, data_eval),
verbose=False)
if ind:
ind.final_weights = model.get_weights()
for loss in history.history['loss']:
if np.isnan(loss):
K.clear_session()
return 0.01, 100
val_loss = history.history['val_loss'][-1]
model = models.Model(input_layer, enc)
indicators = model.predict(data_eval)
# MK test
rho_mk = calc_rho(indicators)
max_rho_mk = 0.01 if np.isnan(max(abs(rho_mk))) else max(abs(rho_mk))
loss = 100 if np.isnan(val_loss) else val_loss
return max_rho_mk, loss