[ENH] Suppress __array_wrap__ warning in numpy 2 for torch and pandas (#1911)

fixes #1764

Adding the comments from the Initial PR
#1855 by @fkiraly


The reason is that `numpy 2` added additional arguments to
`__array_wrap__` which was interacting with
`torch.Tensor.__array_wrap__` or `pandas.Series.__array_wrap__` that did
not have said arguments on earlier versions, in `TorchNormalizer`.

The warning can be avoided by explicit coercion to `torch` prior to
mixed usage - avoiding mixed types across packages entirely.

This also makes the coercions more explicit, enabling future refactors
where we may want to move to more "clean" type assumptions instead of
polymorphic union types.

Author: fnhirwa

pytorch_forecasting/data/encoders.py

@@ -708,17 +708,48 @@ def transform(
             target_scale = self.get_parameters().numpy()[None, :]
         center = target_scale[..., 0]
         scale = target_scale[..., 1]
+
+        if not isinstance(y, torch.Tensor):
+            if isinstance(y, (pd.Series)):
+                index = y.index
+                pandas_dtype = y.dtype
+                y = y.values
+                y_was = "pandas"
+                y = torch.as_tensor(y)
+            elif isinstance(y, np.ndarray):
+                y_was = "numpy"
+                np_dtype = y.dtype
+                try:
+                    y = torch.from_numpy(y)
+                except TypeError:
+                    y = torch.as_tensor(y.astype(np.float32))
+        else:
+            y_was = "torch"
+            torch_dtype = y.dtype
+        if isinstance(center, np.ndarray):
+            center = torch.from_numpy(center)
+        if isinstance(scale, np.ndarray):
+            scale = torch.from_numpy(scale)
         if y.ndim > center.ndim:  # multiple batches -> expand size
             center = center.view(*center.size(), *(1,) * (y.ndim - center.ndim))
             scale = scale.view(*scale.size(), *(1,) * (y.ndim - scale.ndim))
 
-        # transform
-        dtype = y.dtype
         y = (y - center) / scale
-        try:
-            y = y.astype(dtype)
-        except AttributeError:  # torch.Tensor has `.type()` instead of `.astype()`
-            y = y.type(dtype)
+
+        if y_was == "numpy":
+            numpy_data = y.numpy()
+            if np_dtype.kind in "iu" and numpy_data.dtype.kind == "f":
+                # Original was integer, but normalized data is float
+                y = numpy_data.astype(np.float64)
+            else:
+                y = numpy_data.astype(np_dtype)
+        elif y_was == "pandas":
+            numpy_data = y.numpy()
+            if pandas_dtype.kind in "iu" and numpy_data.dtype.kind == "f":
+                pandas_dtype = np.float64
+            y = pd.Series(numpy_data, index=index, dtype=pandas_dtype)
+        else:
+            y = y.type(torch_dtype)
 
         # return with center and scale or without
         if return_norm:
@@ -761,6 +792,10 @@ def __call__(self, data: dict[str, torch.Tensor]) -> torch.Tensor:
         """
         # ensure output dtype matches input dtype
         dtype = data["prediction"].dtype
+        if isinstance(dtype, np.dtype):
+            # convert the array into tensor if it is a numpy array
+            data["prediction"] = torch.as_tensor(data["prediction"])
+            dtype = data["prediction"].dtype
 
         # inverse transformation with tensors
         norm = data["target_scale"]