Fix geometry dtype to respect config.real(np) for NumPy 2.x compatibility

[SimingShan]

Several geometry subclasses (Rectangle, Disk, StarShaped, Triangle, Polygon, Cuboid, GeometryXTime) created NumPy arrays via np.linspace, np.full, np.zeros, np.ones without specifying dtype, defaulting to float64. This caused a dtype mismatch with the neural network (float32 by default), resulting in "mat1 and mat2 must have the same dtype, but got Double and Float" errors in PyTorch with NumPy >= 2.0.

Added dtype=config.real(np) to all affected array-creation calls, consistent with the existing pattern used in geometry_1d.py, geometry_nd.py, and sampler.py.

[echen5503]

Could you provide code to demonstrate the issue?

[SimingShan]

See below
`(base) shansiming@wuzhui:~/project/deepxde$ python test_script.py
Using backend: pytorch
Other supported backends: tensorflow.compat.v1, tensorflow, jax, paddle.
paddle supports more examples now and is recommended.
/home/shansiming/project/deepxde/deepxde/utils/external.py:9: UserWarning: A NumPy version >=1.23.5 and <2.3.0 is required for this version of SciPy (detected version 2.4.3)
import scipy.spatial.distance
NumPy version: 2.4.3
Warning: 50 points required, but 54 points sampled.
uniform_boundary_points dtype: float64
random_boundary_points dtype: float64
Warning: Polygon.uniform_points not implemented. Use random_points instead.
Compiling model...
'compile' took 0.390143 s

Training model...

Traceback (most recent call last):
File "/home/shansiming/project/deepxde/test_script.py", line 23, in
model.train(iterations=100)
File "/home/shansiming/project/deepxde/deepxde/utils/internal.py", line 22, in wrapper
result = f(*args, **kwargs)
^^^^^^^^^^^^^^^^^^
File "/home/shansiming/project/deepxde/deepxde/model.py", line 698, in train
self._test(verbose=verbose)
File "/home/shansiming/project/deepxde/deepxde/model.py", line 891, in _test
) = self._outputs_losses(
^^^^^^^^^^^^^^^^^^^^^
File "/home/shansiming/project/deepxde/deepxde/model.py", line 604, in outputs_losses
outs = outputs_losses(inputs, targets, auxiliary_vars)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/shansiming/project/deepxde/deepxde/model.py", line 354, in outputs_losses_train
return outputs_losses(
^^^^^^^^^^^^^^^
File "/home/shansiming/project/deepxde/deepxde/model.py", line 331, in outputs_losses
outputs = self.net(inputs)
^^^^^^^^^^^^^^^^
File "/home/shansiming/.local/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/shansiming/.local/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/shansiming/project/deepxde/deepxde/nn/pytorch/fnn.py", line 61, in forward
else self.activation(linear(x))
^^^^^^^^^
File "/home/shansiming/.local/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/shansiming/.local/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/shansiming/.local/lib/python3.12/site-packages/torch/nn/modules/linear.py", line 117, in forward
return F.linear(input, self.weight, self.bias)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/shansiming/.local/lib/python3.12/site-packages/torch/utils/_device.py", line 79, in torch_function
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
RuntimeError: mat1 and mat2 must have the same dtype, but got Double and Float`

[SimingShan]

`(base) shansiming@wuzhui:~/project/deepxde$ python test_script.py
Using backend: pytorch
Other supported backends: tensorflow.compat.v1, tensorflow, jax, paddle.
paddle supports more examples now and is recommended.
NumPy version: 1.26.4
Warning: 50 points required, but 54 points sampled.
uniform_boundary_points dtype: float64
random_boundary_points dtype: float32
Warning: Polygon.uniform_points not implemented. Use random_points instead.
Compiling model...
'compile' took 0.384630 s

Training model...

/home/shansiming/.local/lib/python3.12/site-packages/torch/autograd/graph.py:768: UserWarning: Attempting to run cuBLAS, but there was no current CUDA context! Attempting to set the primary context... (Triggered internally at ../aten/src/ATen/cuda/CublasHandlePool.cpp:135.)
return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
Step Train loss Test loss Test metric
0 [1.01e+00, 2.06e-02] [9.91e-01, 2.06e-02] []
100 [6.98e-02, 3.87e-03] [5.69e-02, 3.87e-03] []

Best model at step 100:
train loss: 7.37e-02
test loss: 6.08e-02
test metric: []

'train' took 0.478478 s

Training completed successfully`

[SimingShan]

After the fix
`(base) shansiming@wuzhui:~/project/deepxde$ python test_script.py
Using backend: pytorch
Other supported backends: tensorflow.compat.v1, tensorflow, jax, paddle.
paddle supports more examples now and is recommended.
/home/shansiming/project/deepxde/deepxde/utils/external.py:9: UserWarning: A NumPy version >=1.23.5 and <2.3.0 is required for this version of SciPy (detected version 2.4.3)
import scipy.spatial.distance
NumPy version: 2.4.3
Warning: 50 points required, but 54 points sampled.
uniform_boundary_points dtype: float32
random_boundary_points dtype: float32
Warning: Polygon.uniform_points not implemented. Use random_points instead.
Compiling model...
'compile' took 0.380587 s

Training model...

/home/shansiming/.local/lib/python3.12/site-packages/torch/autograd/graph.py:768: UserWarning: Attempting to run cuBLAS, but there was no current CUDA context! Attempting to set the primary context... (Triggered internally at ../aten/src/ATen/cuda/CublasHandlePool.cpp:135.)
return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
Step Train loss Test loss Test metric
0 [9.77e-01, 4.38e-02] [9.86e-01, 4.38e-02] []
100 [7.47e-02, 3.49e-03] [6.68e-02, 3.49e-03] []

Best model at step 100:
train loss: 7.82e-02
test loss: 7.03e-02
test metric: []

'train' took 0.473906 s

Training completed successfully`

[SimingShan]

Pay attention to the numpy version @echen5503

[echen5503]

Logs are convincing. As a final check, can you provide the test script itself?

[echen5503]

Looks like some files like geometry/pointcloud.py aren't using correct dtype either, maybe good to fix that too