search

privacytrustlab / ml_privacy_meter

Privacy Meter: An open-source library to audit data privacy in statistical and machine learning algorithms.
MIT License
556 stars 99 forks source link

torch.Tensor vs torch.tensor in model.py #109

Closed Menezmic21 closed 11 months ago

Menezmic21 commented 12 months ago

Hello, this is my first issue so please bear with me.

Tldr: Should model.py use torch.tensor() instead of torch.Tensor()?

Explanation: I was adapting the reference_metric tutorial to use Pytorch when I encountered a compatibility issue between the passed tensor and the loss function when creating the Audit object. In particular, the loss function nn.CrossEntropyLoss() required a torch.int64 tensor while it was receiving a torch.float32 tensor. This is despite passing a dtype('int64') numpy array as the value to the y key in the Dataset object used in the InfromationSources in the Audit constructor. I traced back the error and it seems to stem from model.py. I noticed that model.py uses torch.Tensor() (which creates a torch.FloatTensor) instead of torch.tensor() (which infers the dtypeof the tensor automatically). After replacing the code with torch.tensor(), the program ran as expected.

Is there a different way of making this work without editing the source? Or is torch.Tensor() just supposed to be torch.tensor()? Thanks!

changhongyan123 commented 11 months ago

Hi @Menezmic21 , this has been solved in the new version. Please feel free to check it.