good fix: norm_x to float64 to deal with precision issue

Author: sniperyyc

netshare/pre_post_processors/netshare/netshare_pre_post_processor.py

@@ -343,12 +343,12 @@ def _post_process(self, input_folder, output_folder,
         print(f"{self.__class__.__name__}.{inspect.stack()[0][3]}")
 
         # Denormalize the fields (e.g. int to IP, vector to word, etc.)
-        # denormalize_fields(
-        #     config_pre_post_processor=self._config,
-        #     pre_processed_data_folder=pre_processed_data_folder,
-        #     generated_data_folder=input_folder,
-        #     post_processed_data_folder=output_folder
-        # )
+        denormalize_fields(
+            config_pre_post_processor=self._config,
+            pre_processed_data_folder=pre_processed_data_folder,
+            generated_data_folder=input_folder,
+            post_processed_data_folder=output_folder
+        )
 
         # Choose the best generated data across different hyperparameters/checkpoints
         choose_best_model(

netshare/utils/field.py

@@ -64,17 +64,16 @@ def denormalize(self, norm_x):
         if norm_x.shape[-1] != self.dim_x:
             raise ValueError(f"Dimension is {norm_x.shape[-1]}. "
                              f"Expected dimension is {self.dim_x}")
+        norm_x = norm_x.astype(np.float64)  # Convert to float64 for precision
+
         # [0, 1] normalization
         if self.norm_option == Normalization.ZERO_ONE:
-            to_return = np.asarray(
-                norm_x * float(self.max_x - self.min_x) + self.min_x,
-                dtype=np.float64)
+            to_return = norm_x * float(self.max_x - self.min_x) + self.min_x
 
         # [-1, 1] normalization
         elif self.norm_option == Normalization.MINUSONE_ONE:
-            to_return = np.asarray((norm_x + 1) / 2.0 *
-                                   float(self.max_x - self.min_x) + self.min_x,
-                                   dtype=np.float64)
+            to_return = (norm_x + 1) / 2.0 * \
+                float(self.max_x - self.min_x) + self.min_x
 
         else:
             raise Exception("Not valid normalization option!")