Skip to content

neuralbench: --dataset rejects the default dataset and the study names that --help lists #280

Description

@juan-cobos

Description

neuralbench --help shows the datasets you can use for each task. For the EEG image task it shows:

image: Gifford2022Large, Grootswagers2022Human, Xu2024Alljoined, Xu2025Alljoined

None of these names works with --dataset:

  • Gifford2022Large is the task's default dataset. It fails, and the error doesn't even list it as a choice.
  • Grootswagers2022Human also fails. Only the lowercase grootswagers2022human works.

This happens because --help shows each dataset's study name, but --dataset only accepts the name of a file in the task's datasets/ folder, and those file names are lowercase. The default dataset is set in the task's main config.yaml and has no file of its own under datasets/, so --dataset can't select it by name at all.

Steps to reproduce

1. Run neuralbench --help and find the image: line under eeg. It lists Gifford2022Large first.
2. Run neuralbench eeg image --dataset Gifford2022Large
3. It fails with:
ValueError: Unknown dataset 'Gifford2022Large' for eeg/image. Choose from: ['grootswagers2022human', 'xu2024alljoined', 'xu2025alljoined']
4. Run neuralbench eeg image --dataset Grootswagers2022Human. It fails the same way.

Expected: every name that --help lists works with --dataset, and naming the default dataset runs the task with its normal config.

Environment

  • OS: Manjaro Linux
  • Python version: 3.13
  • neuralbench 0.3.1

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions