normalize before upsampling (nans)

modules/deeplearning/srcnn_l1b_l2.py

@@ -225,22 +225,23 @@ class SRCNN:
         for param in data_params:
             idx = params.index(param)
             tmp = input_data[:, idx, 3:131:2, 3:131:2]
-            tmp = resample_2d_linear(x_64, y_64, tmp, t, s)
             tmp = normalize(tmp, param, mean_std_dct, add_noise=add_noise, noise_scale=noise_scale)
+            tmp = resample_2d_linear(x_64, y_64, tmp, t, s)
             data_norm.append(tmp)
         # --------
         idx = params.index('refl_0_65um_nom')
         tmp = input_data[:, idx, 3:131, 3:131]
+        tmp = normalize(tmp, param, mean_std_dct, add_noise=add_noise, noise_scale=noise_scale)
         # tmp = input_data[:, idx, 3:131:2, 3:131:2]
         # tmp = resample_2d_linear(x_64, y_64, tmp, t, s)
-        tmp = normalize(tmp, param, mean_std_dct, add_noise=add_noise, noise_scale=noise_scale)
         data_norm.append(tmp)
         # --------
         tmp = input_data[:, label_idx, 3:131:2, 3:131:2]
-        tmp = np.where(np.isnan(tmp), 0, tmp)
-        tmp = resample_2d_linear(x_64, y_64, tmp, t, s)
         if label_param != 'cloud_fraction':
             tmp = normalize(tmp, param, mean_std_dct, add_noise=add_noise, noise_scale=noise_scale)
+        else:
+            tmp = np.where(np.isnan(tmp), 0, tmp)
+        tmp = resample_2d_linear(x_64, y_64, tmp, t, s)
         data_norm.append(tmp)
         # ---------
         data = np.stack(data_norm, axis=3)
@@ -248,10 +249,11 @@ class SRCNN:
 
         # label = input_data[:, label_idx, 3:131:2, 3:131:2]
         label = input_data[:, label_idx, 3:131, 3:131]
-        label = np.where(np.isnan(label), 0, label)
-        label = np.expand_dims(label, axis=3)
         if label_param != 'cloud_fraction':
             label = normalize(label, label_param, mean_std_dct)
+        else:
+            label = np.where(np.isnan(label), 0, label)
+        label = np.expand_dims(label, axis=3)
 
         data = data.astype(np.float32)
         label = label.astype(np.float32)
@@ -692,16 +694,16 @@ def run_evaluate_static(in_file, out_file, ckpt_dir):
     x_up = np.arange(0, lenx, 0.5)
     y_up = np.arange(0, leny, 0.5)
 
-    grd_a = resample_2d_linear_one(x, y, grd_a, x_up, y_up)
     grd_a = normalize(grd_a, 'temp_11_0um_nom', mean_std_dct)
+    grd_a = resample_2d_linear_one(x, y, grd_a, x_up, y_up)
 
     grd_b = normalize(grd_b, 'refl_0_65um_nom', mean_std_dct)
 
-    grd_c = resample_2d_linear_one(x, y, grd_c, x_up, y_up)
     if label_param == 'cloud_fraction':
         grd_c = np.where(np.isnan(grd_c), 0, grd_c)
     else:
         grd_c = normalize(grd_c, label_param, mean_std_dct)
+    grd_c = resample_2d_linear_one(x, y, grd_c, x_up, y_up)
 
     data = np.stack([grd_a, grd_b, grd_c], axis=2)
     data = np.expand_dims(data, axis=0)