enh: more datasets and augmentation #61
Conversation
stephane-rivaud
commented
Feb 13, 2025
stephane-rivaud
commented
- refactor SinDataset
- adding FashionMNIST and SVHN
- support for AutoAugment and RandAugment
- refactor SinDataset - adding FashionMNIST and SVHN - support for AutoAugment and RandAugment
- refactor SinDataset - adding FashionMNIST and SVHN - support for AutoAugment and RandAugment
Codecov ReportAll modified and coverable lines are covered by tests ✅
Flags with carried forward coverage won't be shown. Click here to find out more.
|
- refactor SinDataset - adding FashionMNIST and SVHN - support for AutoAugment and RandAugment
src/gromo/tools.py
Outdated
| @@ -6,7 +6,7 @@ | |||
| def sqrt_inverse_matrix_semi_positive( | |||
| matrix: torch.Tensor, | |||
| threshold: float = 1e-5, | |||
| preferred_linalg_library: None | str = "magma", | |||
| preferred_linalg_library: None | str = None, | |||
There was a problem hiding this comment.
Ok to apply this change as we could assume new hardware and software environment for code execution. We could explicit somewhere that older cuda version may be subject to the bad SVD values.
There was a problem hiding this comment.
Actually, I removed it because I wanted to try out Stella's solution first.
I always need to do this to test things locally, but we "could" adopt this quick fix if Stella's solution does not work properly.
(So far it does not work properly for me but I need to check if this comes from my setup)
There was a problem hiding this comment.
With Pull request #60 I made the config file non-shared so you can create your own and test if it works
|
Outdated |
