def __init__(self, params_dict):
"""Constructor
Args:
params_dict (dict): Contains training/model parameters
"""
self.save_dir = file_utils.make_save_directory("cWGAN")
self.num_critic_iters = params_dict["num_critic_iters"]
self.batch_size = params_dict["batch_size"]
clip_value = params_dict["clip_value"]
noise_dims = params_dict["noise_dims"]
gen_lr = params_dict["gen_lr"]
critic_lr = params_dict["critic_lr"]
gp_weight = params_dict["gp_weight"]
self.model = cWGAN_mnist(clip_value, noise_dims, gen_lr, critic_lr,
gp_weight)
self.data = data_utils.load_mnist_data()
self.epochs = params_dict["epochs"]
self.num_training_examples = len(self.data[0])
self.weight_saving_interval = params_dict["weight_saving_interval"]
self.critic_losses = []
self.generator_losses = []
self.wass_estimates = []
def __init__(self, params_dict):
"""Constructor
Args:
params_dict (dict): Contains training/model parameters
"""
self.save_dir = file_utils.make_save_directory("cWGAN")
self.num_critic_iters = params_dict["num_critic_iters"]
self.batch_size = params_dict["batch_size"]
gen_lr = params_dict["gen_lr"]
critic_lr = params_dict["critic_lr"]
optimizer = params_dict["optimizer"]
noise_dims = params_dict["noise_dims"]
gp_weight = params_dict["gp_weight"]
load_previous = params_dict["load_previous"]
weights_path = params_dict["weights_path"]
iteration = params_dict["iteration"]
self.model = cWGAN(noise_dims, optimizer, gen_lr, critic_lr, gp_weight,
load_previous, weights_path, iteration)
data_path = params_dict["data_path"]
if params_dict["data_type"] == "angular":
if params_dict["data_scaling"] == "inverse":
self.data = data_utils.load_jet_data_inverse_scaling(data_path)
elif params_dict["data_scaling"] == "log":
self.data = data_utils.load_jet_data_log_scaling(data_path)
elif params_dict["data_scaling"] == "minmax":
self.data = data_utils.load_jet_data(data_path)
else:
print("There was an error loading the data")
print(
"Please specify a data scaling scheme in your configuration file: The options are: inverse, log, and minmax"
)
sys.exit(1)
elif params_dict["data_type"] == "cartesian":
self.data = data_utils.load_jet_data_log_scaling_cartesian(
data_path)
else:
print("There was an error loading the data")
print(
"Please specify the form of the data your are training on, either:"
)
print(" angular: Four-momenta in form (Pt, Eta, Phi, E)")
print(" cartesian: Four-momenta in form (Px, Py, Pz, E)")
sys.exit(1)
self.epochs = params_dict["epochs"]
self.num_training_examples = len(self.data[0])
self.weight_saving_interval = params_dict["weight_saving_interval"]
self.critic_losses = []
self.generator_losses = []
self.wass_estimates = []
def __init__(self, params_dict):
"""Constructor
Args:
"""
self.save_dir = file_utils.make_save_directory("Classifier")
self.model = build_model()
self.model.compile(
optimizer=tf.keras.optimizers.Adam(
learning_rate=params_dict['lr']),
loss=tf.keras.losses.BinaryCrossentropy(),
metrics=[tf.keras.metrics.BinaryAccuracy()],
)
self.batch_size = params_dict["batch_size"]
self.data = data_utils.load_classifier_data(params_dict["data_path"])
self.checkpoint_path = os.path.join(self.save_dir, "training/cp.cpkt")
self.epochs = params_dict["epochs"]