physai 2.5.2__tar.gz → 2.5.7__tar.gz

{physai-2.5.2/src/physai.egg-info → physai-2.5.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: physai
-Version: 2.5.2
+Version: 2.5.7
 Summary: Physics-Informed Neural Network (PINN) library for solving and animating ODEs/PDEs with visualization and training tools
 Author-email: Mankrit Singh <whataninfinity@gmail.com>
 License: MIT

{physai-2.5.2 → physai-2.5.7}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "physai"
-version = "2.5.2"
+version = "2.5.7"
 description = "Physics-Informed Neural Network (PINN) library for solving and animating ODEs/PDEs with visualization and training tools"
 authors = [{name = "Mankrit Singh", email = "whataninfinity@gmail.com"}]
 readme = "README.md"

{physai-2.5.2 → physai-2.5.7}/src/physai/pinn.py

@@ -4,7 +4,11 @@ import torch.optim as optim
 from torch.amp import autocast
 from torch.amp import grad_scaler as GradScaler
 import warnings
-warnings.filterwarnings('ignore', category=torch.exceptions.ComplexHalfWarning)
+try:
+    warnings.filterwarnings('ignore', category=torch.exceptions.ComplexHalfWarning)
+except AttributeError:
+    # Ignore if the warning class doesn't exist (older PyTorch)
+    pass
 
 class PINN(nn.Module):
     """Physics-Informed Neural Network (PINN) with advanced optimization features."""
@@ -53,7 +57,9 @@ class PINN(nn.Module):
                 imag = output[..., 1]
             else:
                 raise ValueError(f"Expected output shape with last dim 2 for complex, got {output.shape}")
-            return torch.complex(real, imag, dtype=torch.complex64)
+            complex_output = torch.complex(real, imag)  # PyTorch infers dtype from inputs
+            complex_output = complex_output.to(torch.complex64)  # optional, ensure 32-bit
+            return complex_output
         return output
 
     def physics_loss(self, x, physics_fn):

{physai-2.5.2 → physai-2.5.7/src/physai.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: physai
-Version: 2.5.2
+Version: 2.5.7
 Summary: Physics-Informed Neural Network (PINN) library for solving and animating ODEs/PDEs with visualization and training tools
 Author-email: Mankrit Singh <whataninfinity@gmail.com>
 License: MIT