snapshot...

modules/deeplearning/icing_fcn.py

@@ -20,8 +20,8 @@ if NumClasses == 2:
 else:
     NumLogits = NumClasses
 
-BATCH_SIZE = 256
-NUM_EPOCHS = 60
+BATCH_SIZE = 128
+NUM_EPOCHS = 50
 
 TRACK_MOVING_AVERAGE = False
 EARLY_STOP = True
@@ -584,18 +584,18 @@ class IcingIntensityFCN:
         # activation = tf.nn.elu
         activation = tf.nn.leaky_relu
 
-        num_filters = len(self.train_params) * 16
+        num_filters = len(self.train_params) * 4
 
         input_2d = self.inputs[0]
 
         if NOISE_TRAINING:
             conv = tf.keras.layers.GaussianNoise(stddev=NOISE_STDDEV)(input_2d)
 
-        conv = tf.keras.layers.Conv2D(num_filters, kernel_size=5, strides=1, padding=padding, activation=activation)(conv)
+        conv = tf.keras.layers.Conv2D(num_filters, kernel_size=3, strides=1, padding=padding, activation=activation)(conv)
         print(conv.shape)
         skip = conv
 
-        conv = tf.keras.layers.Conv2D(num_filters, kernel_size=5, strides=1, padding=padding, activation=activation)(conv)
+        conv = tf.keras.layers.Conv2D(num_filters, kernel_size=3, strides=1, padding=padding, activation=activation)(conv)
         conv = tf.keras.layers.MaxPool2D(padding=padding)(conv)
         if do_drop_out:
             conv = tf.keras.layers.Dropout(drop_rate)(conv)
@@ -694,11 +694,11 @@ class IcingIntensityFCN:
 
         conv = build_residual_block_1x1(conv, num_filters, activation, 'Residual_Block_2', padding=padding)
 
-        conv = build_residual_block_1x1(conv, num_filters, activation, 'Residual_Block_3', padding=padding)
+        # conv = build_residual_block_1x1(conv, num_filters, activation, 'Residual_Block_3', padding=padding)
 
-        conv = build_residual_block_1x1(conv, num_filters, activation, 'Residual_Block_4', padding=padding)
+        # conv = build_residual_block_1x1(conv, num_filters, activation, 'Residual_Block_4', padding=padding)
 
-        conv = build_residual_block_1x1(conv, num_filters, activation, 'Residual_Block_5', padding=padding)
+        # conv = build_residual_block_1x1(conv, num_filters, activation, 'Residual_Block_5', padding=padding)
 
         print(conv.shape)