RFC: in search of the perfect sample data

After revising lots of tutorials, I’ve formed some opinions about what the ideal sample dataset would look like. Here are some thoughts; please feel free to edit / add to the list; maybe if we reach some consensus we can find a dataset that meets the criteria and could replace sample dataset in most/all of the tutorials.

• probably should be neuromag and include fine cal, crosstalk, and a cHPI .pos file (current sample data lacks .pos)
• has mag, grad, EEG, EOG, and ECG (current sample data lacks ECG)
• has multiple subjects (useful for demoing corrmap, stats, …)
• multiple runs per subject maybe? (for things like concatenating raws)
• a single Raw recording is ~90 seconds max, and/or has a low-ish acquisition sampling frequency (so we’re not constantly cropping to save memory)
• raw file does not already contain projectors; separate empty-room file included
• I think the task (L/R aud/vis + catch trials) is pretty good as a demo, but might be nice to have reaction time data included (for things like the epochs image overlay). I’m certainly open to other ideas about the paradigm though.
• EEG channel names follow 10-20 / 10-05 naming conventions

[larsoner]

During annual maintenance (or sometime) an Elekta tech will make an empty room recording, compute projectors, and set these as acquisition projectors that will be used (until the next time they are updated, which might be never, might be in a year, etc.). The goal is to make it so that huge deflections due to env noise do not make it impossible to see real activity during acquisition of subject data.

Empty-room projectors computed from empty room recordings the same day are better because they give more accurate projectors.

[jona-sassenhagen]

A continuous variable (like RT) would allow demoing a lot of things much easier indeed.