dde_config.py

def dde_NN_config():
	#draft 1
	config = {}
	config['batch_size'] = 128
	config['input_dim'] = 1722 #343 #1722 #343 #2161 #1937
	config['batch_first'] = True
	config['num_class'] = 2
	config['LR'] = 1e-3
	config['train_epoch'] = 3
	config['pretrain_epoch'] = 1

	config['recon_threshold'] = 0.0005 # change later

	config['encode_fc1_dim'] = 500  # encoder fc1
	config['encode_fc2_dim'] = 50  # encoder fc2
	config['decode_fc1_dim'] = 500  # decoder fc1
	config['decode_fc2_dim'] = config['input_dim']  # decoder reconstruction
	config['predict_dim'] = 1024 # for every layer
	config['predict_out_dim'] = 1 # predictor out
	config['lambda1'] = 1e-2  # L1 regularization coefficient
	config['lambda2'] = 1e-1  # L2 regulatization coefficient
	config['lambda3'] = 1e-5  # L2 regulatization coefficient
	config['reconstruction_coefficient'] = 1e-1  # 1e-2
	config['projection_coefficient'] = 1e-1  # 1e-2
	config['magnify_factor'] = 100
	return config