Skip to content

Idea: optional "evaluation provenance" for datasets (useful, or out of scope)? #391

Description

@YG-paaleee

Hi! I'm a CS student getting into BCI, and I recently built a small open-source tool called EEGauge (https://github.com/YG-paaleee/eegauge) that reads EEGDash metadata. Your contributing guide says to open an issue before starting on a feature, so I wanted to float an idea and get your read before building anything.

EEGDash already handles finding, validating, and loading datasets really well. The part I keep wishing existed is the step after that: how a dataset actually gets evaluated, and whether a result should be trusted. Two concrete examples: flagging when a dataset has multiple sessions or runs per subject (so a naive epoch-level shuffle would leak), and recording the exact cohort that produced a benchmark so someone can reproduce it. EEGauge already spits out a small "evaluation provenance" file like that (here's one for ds002718: https://github.com/YG-paaleee/eegauge/blob/main/examples/ds002718.provenance.json), and it seemed like it might complement your dataset pages instead of overlapping with them.

Mostly I just wanted to ask before doing anything: is this something you'd find useful on the EEGDash side, or is it out of scope and better left to downstream tools like mine? If you're open to it, I'd rather hear how you'd want it shaped (which fields are stable, where it would even fit) than guess and send a PR you didn't ask for.

Either way, thanks for EEGDash. It's been genuinely nice to build on.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions