search

LMZimmer / Auto-PyTorch_refactor

Apache License 2.0
0 stars 2 forks source link

Inconsistency in usage of train_tensors (Image Dataset) #48

Open ravinkohli opened 3 years ago

ravinkohli commented 3 years ago

So, currently the base dataset is expecting train_tensors to be a Tuple(X: List,y: List) i.e, it is a tuple containing a list of X's and a list of targets. However, the image dataset class uses torch.utils.dataset which has training data as List(Tuple(X, y)), i.e, it is a list of tuples where the tuples contain image and the corresponding target.

bastiscode commented 3 years ago

I think the issue you are refering to is addressed in current base dataset in the data_dev branch (https://github.com/LMZimmer/Auto-PyTorch_refactor/blob/data_dev/autoPyTorch/datasets/base_dataset.py). There the input for the base dataset can either be a torch.utils.Dataset or a tuple of indexable objects.

ravinkohli commented 3 years ago

Okay, I'll wait for this to be merged as its different on the development branch. Currently, I have skipped the image dataset test, when this is merged could you run the ImageDataset test and close this issue?