snapshot...

modules/deeplearning/icing_cnn.py

@@ -13,6 +13,7 @@ from icing.pirep_goes import normalize, make_for_full_domain_predict
 
 LOG_DEVICE_PLACEMENT = False
 
+# Manual (data, label) caching, but has been replaced with tf.data.dataset.cache()
 CACHE_DATA_IN_MEM = False
 
 PROC_BATCH_SIZE = 4096
@@ -37,8 +38,8 @@ NOISE_TRAINING = False
 
 img_width = 16
 
-mean_std_file = homedir+'data/icing/mean_std_no_ice.pkl'
-# mean_std_file = homedir+'data/icing/mean_std_l1b_no_ice.pkl'
+# mean_std_file = homedir+'data/icing/mean_std_no_ice.pkl'
+mean_std_file = homedir+'data/icing/mean_std_l1b_no_ice.pkl'
 f = open(mean_std_file, 'rb')
 mean_std_dct = pickle.load(f)
 f.close()
@@ -47,13 +48,13 @@ f.close()
 # train_params = ['cld_height_acha', 'cld_geo_thick', 'cld_temp_acha', 'cld_press_acha', 'supercooled_cloud_fraction',
 #                 'cld_emiss_acha', 'conv_cloud_fraction', 'cld_reff_acha', 'cld_opd_acha']
 # -- DAY L2 -------------
-train_params = ['cld_height_acha', 'cld_geo_thick', 'cld_temp_acha', 'cld_press_acha', 'supercooled_cloud_fraction',
-#                'cld_emiss_acha', 'conv_cloud_fraction', 'cld_reff_dcomp', 'cld_opd_dcomp', 'cld_cwp_dcomp', 'iwc_dcomp', 'lwc_dcomp']
-                'cld_emiss_acha', 'conv_cloud_fraction', 'cld_reff_dcomp', 'cld_opd_dcomp', 'iwc_dcomp', 'lwc_dcomp']
+#train_params = ['cld_height_acha', 'cld_geo_thick', 'cld_temp_acha', 'cld_press_acha', 'supercooled_cloud_fraction',
+##                'cld_emiss_acha', 'conv_cloud_fraction', 'cld_reff_dcomp', 'cld_opd_dcomp', 'cld_cwp_dcomp', 'iwc_dcomp', 'lwc_dcomp']
+#                'cld_emiss_acha', 'conv_cloud_fraction', 'cld_reff_dcomp', 'cld_opd_dcomp', 'iwc_dcomp', 'lwc_dcomp']
 # -- DAY L1B --------------------------------
-# train_params = ['temp_10_4um_nom', 'temp_11_0um_nom', 'temp_12_0um_nom', 'temp_13_3um_nom', 'temp_3_75um_nom',
-#                 'temp_6_2um_nom', 'temp_6_7um_nom', 'temp_7_3um_nom', 'temp_8_5um_nom', 'temp_9_7um_nom',
-#                 'refl_0_47um_nom', 'refl_0_65um_nom', 'refl_0_86um_nom', 'refl_1_38um_nom', 'refl_1_60um_nom']
+train_params = ['temp_10_4um_nom', 'temp_11_0um_nom', 'temp_12_0um_nom', 'temp_13_3um_nom', 'temp_3_75um_nom',
+                'temp_6_2um_nom', 'temp_6_7um_nom', 'temp_7_3um_nom', 'temp_8_5um_nom', 'temp_9_7um_nom',
+                'refl_0_47um_nom', 'refl_0_65um_nom', 'refl_0_86um_nom', 'refl_1_38um_nom', 'refl_1_60um_nom']
 # -- NIGHT L1B -------------------------------
 # train_params = ['temp_10_4um_nom', 'temp_11_0um_nom', 'temp_12_0um_nom', 'temp_13_3um_nom', 'temp_3_75um_nom',
 #                 'temp_6_2um_nom', 'temp_6_7um_nom', 'temp_7_3um_nom', 'temp_8_5um_nom', 'temp_9_7um_nom']
@@ -223,9 +224,8 @@ class IcingIntensityNN:
         if not is_training:
             h5f = self.h5f_tst
 
-        key = frozenset(idxs)
-
         if CACHE_DATA_IN_MEM:
+            key = frozenset(idxs)
             tup = self.in_mem_data_cache.get(key)
             if tup is not None:
                 return tup[0], tup[1]
@@ -436,7 +436,7 @@ class IcingIntensityNN:
         momentum = 0.99
 
         # num_filters = 16
-        num_filters = 24
+        num_filters = 30
 
         conv = tf.keras.layers.Conv2D(num_filters, 5, strides=[1, 1], padding=padding, activation=activation)(self.inputs[0])
         conv = tf.keras.layers.MaxPool2D(padding=padding)(conv)