[Fix] Fix weight dtype and bug in symbolic (#709)

* fix dtype of weight to avoid unnecessary type promotion

* fix bug in symbolic

Author: HydrogenSulfate

examples/hpinns/functions.py

@@ -100,8 +100,8 @@ def init_lambda(output_dict: Dict[str, paddle.Tensor], bound: int):
     """
     global lambda_re, lambda_im, loss_weight
     x, y = output_dict["x"], output_dict["y"]
-    lambda_re = np.zeros((len(x[bound:]), 1))
-    lambda_im = np.zeros((len(y[bound:]), 1))
+    lambda_re = np.zeros((len(x[bound:]), 1), paddle.get_default_dtype())
+    lambda_im = np.zeros((len(y[bound:]), 1), paddle.get_default_dtype())
     # loss_weight: [PDE loss 1, PDE loss 2, Lagrangian loss 1, Lagrangian loss 2, objective loss]
     if train_mode == "aug_lag":
         loss_weight = [0.5 * mu] * 2 + [1.0, 1.0] + [1.0]

ppsci/data/dataset/array_dataset.py

@@ -108,7 +108,10 @@ def __init__(
         self.input_keys = tuple(input.keys())
         self.label_keys = tuple(self.label.keys())
         self.weight = (
-            {key: paddle.to_tensor(value) for key, value in weight.items()}
+            {
+                key: paddle.to_tensor(value, paddle.get_default_dtype())
+                for key, value in weight.items()
+            }
             if weight is not None
             else None
         )

ppsci/data/dataset/trphysx_dataset.py

@@ -130,7 +130,8 @@ def __getitem__(self, i):
 
         weight_shape = [1] * len(data_item.shape)
         weight_item = {
-            key: np.full(weight_shape, value) for key, value in self.weight_dict.items()
+            key: np.full(weight_shape, value, paddle.get_default_dtype())
+            for key, value in self.weight_dict.items()
         }
         return (input_item, label_item, weight_item)
 
@@ -307,6 +308,7 @@ def __getitem__(self, i):
                 label_item[self.label_keys[1]] = data_item[1:, :]
         weight_shape = [1] * len(data_item.shape)
         weight_item = {
-            key: np.full(weight_shape, value) for key, value in self.weight_dict.items()
+            key: np.full(weight_shape, value, paddle.get_default_dtype())
+            for key, value in self.weight_dict.items()
         }
         return (input_item, label_item, weight_item)

ppsci/utils/symbolic.py

@@ -256,8 +256,8 @@ def _minimum_operator_func(self, data_dict: DATA_DICT) -> DATA_DICT:
         )
         for i in range(2, len(self.childs)):
             data_dict[self.key] = paddle.minimum(
-                data_dict[data_dict[self.key]],
-                data_dict[data_dict[self.childs[i]]],
+                data_dict[self.key],
+                data_dict[self.childs[i]],
             )
         return data_dict
 
@@ -267,8 +267,8 @@ def _maximum_operator_func(self, data_dict: DATA_DICT) -> DATA_DICT:
         )
         for i in range(2, len(self.childs)):
             data_dict[self.key] = paddle.maximum(
-                data_dict[data_dict[self.key]],
-                data_dict[data_dict[self.childs[i]]],
+                data_dict[self.key],
+                data_dict[self.childs[i]],
             )
         return data_dict