When I'm debugging with PyCharm, I'd like for the debugger to display the shapes of my NumPy arrays / Jax arrays / PyTorch Tensors. Instead, I see their values:
Is there a way to configure PyCharm's debugger so that the shapes of these multi-dimensional arrays is displayed?
I achieve this by hacking member function torch.Tensor.__repr__, because PyCharm debugger calls __repr__ to get string representation of the object.
import torch
old_repr = torch.Tensor.__repr__
def tensor_info(tensor):
return repr(tensor.shape)[6:] + ' ' + repr(tensor.dtype)[6:] + '@' + str(tensor.device) + '\n' + old_repr(tensor)
torch.Tensor.__repr__ = tensor_info
In PyCharm debugger, you will see representation like:
>>>print(torch.ones(3,3))
Size([3, 3]) float32@cpu
tensor([[1., 1., 1.],
[1., 1., 1.],
[1., 1., 1.]])
Snapshot of my variable inspector:
Debugger just shows the objects in memory. So if you create an object representing shape, you should see it in the debugger.
For example, for numpy it's numpy.array().shape
np.array().shape is displayed next to the object without needing to "open" the object in the debugger console