Hi,
To remove aforementioned dependency, PyTorch built-in method has been used. Although, it has a small bug which can be resolved by detaching tensors with grad as the output of torch.searchsorted is int so it does not have gradients.
Also, the readme has been updated regarding this change, and a typo has been fixed.
Furthermore, I added my experience with model for training on fern dataset (time wise).
Hi, To remove aforementioned dependency, PyTorch built-in method has been used. Although, it has a small bug which can be resolved by detaching tensors with grad as the output of
torch.searchsorted
isint
so it does not have gradients.Also, the readme has been updated regarding this change, and a typo has been fixed.
Furthermore, I added my experience with model for training on
fern
dataset (time wise).