settings.py

class gb(object):
    # Paths
	LOGDIR = '/TRAIN/model/'
	outpath = '/TEST/val/'
	impath = '/Paths/Train/'
	testpath = '/Paths/Test/'

	# Set variables
	learning_rate = 0.0001
	lr_decay = [3,6,9,12,15]
	num_epochs = 15
	sub_epochs = 20
	batch_size = 10
	factor = 100 # Number of patients per epoch
	patches_per_patient = 20
	num_classes = 5 
	num_ch = 4
	ends = ["Channels_flair.txt", "Channels_t1.txt", "Channels_t1c.txt", "Channels_t2.txt", "GtLabels.txt"] # Last one: annotations
	train_patches = 40 # Size Train patches
	test_patches = 20 # Size Test patches
	num_cases = 285 # Number of patients in the dataset

	# Architecture parameters
	num_paths = 4 # First will be normal, the others will receive a downsampled image
	downsample = [3, 6, 8] # Factors to downsample inputs
	fc_layers = 2 # Number of fully connected layers
	fc_drop = 0.5 # The probability that each element is kept
	convolutions = [num_ch, 30, 30, 40, 40, 40, 40] # initial convolution + layers [block]
	conv_drop = 0.98