snapshot...

modules/deeplearning/icing_fcn.py

@@ -1198,11 +1198,7 @@ def run_average_models(ckpt_dir_s_path, day_night='NIGHT', l1b_andor_l2='BOTH',
         for k, w in enumerate(m):
             model_lyrs[k].append(w)
     for lyr in model_lyrs:
-        nda = np.stack(lyr, axis=-1)
-        print(nda.shape)
-        avg = np.mean(nda, axis=-1)
-        print(avg.shape)
-        avg_model_weights.append(nda)
+        avg_model_weights.append(np.mean(np.stack(lyr, axis=-1), axis=-1))
 
     # -- Make a new model for the averaged weights
     new_model = IcingIntensityFCN(day_night=day_night, l1b_or_l2=l1b_andor_l2, use_flight_altitude=use_flight_altitude)
@@ -1211,13 +1207,13 @@ def run_average_models(ckpt_dir_s_path, day_night='NIGHT', l1b_andor_l2='BOTH',
     new_model.build_evaluation()
 
     # -- Save the averaged weights to a new the model
-    # if not os.path.exists(modeldir):
-    #     os.mkdir(modeldir)
-    # ckpt = tf.train.Checkpoint(step=tf.Variable(1), model=new_model.model)
-    # ckpt_manager = tf.train.CheckpointManager(ckpt, modeldir, max_to_keep=3)
-    #
-    # new_model.model.set_weights(avg_weights)
-    # ckpt_manager.save()
+    if not os.path.exists(modeldir):
+        os.mkdir(modeldir)
+    ckpt = tf.train.Checkpoint(step=tf.Variable(1), model=new_model.model)
+    ckpt_manager = tf.train.CheckpointManager(ckpt, modeldir, max_to_keep=3)
+
+    new_model.model.set_weights(avg_model_weights)
+    ckpt_manager.save()
 
     return