Mainak Jas

I have this on my agenda too, great minds think alike ;-) I think this would be great to have. I have some private code and we can exchange ideas....

> Therefore, detecting one threshold per location would be best, and shouldn't be a problem with your method, right? correct, but you do need a mix of good and bad...

@chapochn sorry for the delayed response. Is there a public dataset that could be used to try this on? Maybe something on openneuro?

Hi @wkruijne ! Indeed, this would be an interesting usecase of autoreject that we have considered before. However, it has not been thoroughly tested by me. I would be curious...

can you share a part of your data somehow so I can investigate?

I will need your full data and script to reproduce. Can you share on dropbox? You can use `epochs.save`

sorry I need to update the email address. Could you send to mainakjas [at] gmail [dot] com

@henrikroehr I did manage to get around to this today. Here is the result of my investigations:  code: https://gist.github.com/jasmainak/05dc7c6d26592297d39a4fa25fb21d88 The crux of the problem is that cross-validation in autoreject...

I tried `cv=1` but sklearn was complaining. Do you know if it's allowed?

excellent! Yes that's what I was looking for. In that case, we can add a `cv` argument to `compute_thresholds` that by default does `cv=10` but can also be `cv=NoSplitter()`. So...