TorchSnooper v0.8

• Add support for named tensor.

TorchSnooper v0.7.1

Remove f-string to improve compatibility on older python 3.

TorchSnooper v0.7

Improve support for integral and boolean tensors: On some versions of PyTorch torch.isinf and torch.isnan does not support integral or boolean tensor well, this leads to REPR FAIL in TorchSnooper. This problem is fixed in this version.

TorchSnooper v0.6.1

The output of snoop verbose mode is improved

TorchSnooper v0.6

Add snoop support.

TorchSnooper v0.5

TorchSnooper now reports nan and inf when detected.

TorchSnooper v0.4.1

Add a long description as README.md

TorchSnooper v0.4

Add simple support for the representation of numpy.ndarray.

TorchSnooper v0.3

Containers are now represented recursively.

Example code:

import torch
import torchsnooper

@torchsnooper.snoop()
def f():
    return [{'key': torch.zeros(5, 6, 7)}]

f()

Output from TorchSnooper v0.2

14:17:49.646680 call         5 def f():
14:17:49.646778 line         6     return [{'key': torch.zeros(5, 6, 7)}]
14:17:49.646973 return       6     return [{'key': torch.zeros(5, 6, 7)}]
Return value:.. [{'key': tensor([[[0., 0., 0., 0., 0., 0., 0.], ...., 0.],         [0., 0., 0., 0., 0., 0., 0.]]])}]

Output from TorchSnooper v0.3

14:16:33.205156 call         5 def f():
14:16:33.205261 line         6     return [{'key': torch.zeros(5, 6, 7)}]
14:16:33.205427 return       6     return [{'key': torch.zeros(5, 6, 7)}]
Return value:.. [{'key': tensor<(5, 6, 7), float32, cpu>}]

TorchSnooper v0.2

Now depend on the new official release of PySnooper, not my custom one.