Skip to content

Discrepancy in BCEWithLogitsLoss and ClassNLLLoss #2180

@shubhamjain0594

Description

@shubhamjain0594

In BCEWithLogitsLoss, weights are unnormalized and used, while in ClassNLLLoss weights are normalized and used.

Some discussion around this can be found here. Is this discrepancy expected? Else if this is an issue I can send a PR to resolve this.

Metadata

Metadata

Assignees

No one assigned

    Labels

    module: cpuCPU specific problem (e.g., perf, algorithm)module: cudaRelated to torch.cuda, and CUDA support in generalmodule: lossProblem is related to loss functiontriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions