Update sample.cfg

sample.cfg

@@ -13,45 +13,59 @@ gap_location=250
 ; z-slice index relative to dataset
 
 
-
 ; Only necessary if global.train=true
 [train]
 num_epochs=100
 num_minibatch=64
-minibatch_size=4
+minibatch_size=2
 instance_noise=false
 
+; Instance Noise doesn't really help, this is always set to false
 ; Only necessary if instance_noise=true
 instance_noise_std_dev=0
 anneal_noise=30
 
 ; Only needed if pretrain=false
 pretrained_model=/path/to/model.h5
 generator_learning_rate=1e-4
+
 generator_optimizer=adam
+; Currently not used, generator optimizer is always Adam
+
 generator_mask_size=10
+; Total thickness of mask size, diameter not radius
+
 penalty_learning_rate=1e-4
+; For applying the "penalty" for deviating outside of mask area
+
 discriminator_learning_rate=1e-5
+
 discriminator_optimizer=adam
+; Currently not used, discriminator optimizer is always Adam
+
 discriminator_architecture=default
+; See bottom of models.py for valid options (Currently "a", "b", "c", "default")
 
 ; Anything in this format gets passed to the constructor for that architecture
 discriminator_arg_[argname]=[value]
+; No Square Brackets
 
-discriminator_batchnorm=true
-
+;e.g. discriminator_arg_init_filters=19
+;discriminator_arg_batch_norm=false
 
 
 ; Only necessary if global.pretrain=true
 [pretrain]
 num_epochs=100
 num_minibatch=64
-minibatch_size=4
+minibatch_size=2
+
 generator_architecture=default
+; See bottom of models.py for valid options (Currently "a", "b", "default")
+
+generator_learning_rate=1e-4
 
 ; Anything in this format gets passed to the constructor for that architecture
 generator_arg_[argname]=[value]
-
-generator_learning_rate=1e-4
-generator_regularization=0.0
-generator_batchnorm=true
+;e.g. generator_arg_regularization=0.0
+;generator_arg_batch_norm=true