Update restful/onnx_utilities.py

restful/onnx_utilities.py

@@ -30,7 +30,12 @@ class Utilities:
         days: int, sequence_length: int, model_name: str) -> tuple:
 
         model_path = os.path.join(self.model_path, f'{model_name}.onnx')
-        session = ort.InferenceSession(model_path)
+        # session = ort.InferenceSession(model_path)
+        try:
+            session = ort.InferenceSession(model_path)
+        except Exception as e:
+            print("ONNX model load error:", e)
+            return [], []
         input_name = session.get_inputs()[0].name
 
         dataframe_path = os.path.join(self.posttrained_path, f'{model_name}.csv')
@@ -59,20 +64,38 @@ class Utilities:
 
         #     lst_seq = np.roll(lst_seq, shift=-1, axis=1)
         #     lst_seq[:, -1, -1] = predicted[0][0][0]
+
+
+        
+        # for _ in range(days):
+        #     predicted = session.run(None, {input_name: lst_seq.astype(np.float32)})[0]
+        
+        #     value = np.array(predicted).flatten()[0]
+        #     denorm_price = (value * (max_close - min_close)) + min_close
+        
+        #     # last_date += pd.Timedelta(days=1)
+        #     last_date = pd.to_datetime(last_date) + pd.Timedelta(days=1)
+        #     # predicted_prices[last_date.strftime('%Y-%m-%d')] = float(denorm_price)
+        #     predicted_prices[last_date] = self.truncate_2_decimal(denorm_price)
+        
+        #     lst_seq = np.roll(lst_seq, shift=-1, axis=1)
+        #     lst_seq[:, -1, -1] = value
+        
         for _ in range(days):
             predicted = session.run(None, {input_name: lst_seq.astype(np.float32)})[0]
-        
             value = np.array(predicted).flatten()[0]
-            denorm_price = (value * (max_close - min_close)) + min_close
-        
-            # last_date += pd.Timedelta(days=1)
+            if np.isnan(value):
+                continue
+            denorm_price = self.denormalization(value, min_close, max_close)
+            if np.isnan(denorm_price):
+                continue
             last_date = pd.to_datetime(last_date) + pd.Timedelta(days=1)
-            # predicted_prices[last_date.strftime('%Y-%m-%d')] = float(denorm_price)
             predicted_prices[last_date] = self.truncate_2_decimal(denorm_price)
-        
             lst_seq = np.roll(lst_seq, shift=-1, axis=1)
             lst_seq[:, -1, -1] = value
 
+            
+
         # predictions = [
         #     {'date': date.strftime('%Y-%m-%d'), 'price': float(price)}
         #     for date, price in predicted_prices.items()