Mainak Jas

This could be an interesting hack for a weekend ;)

I'll get to these later this week. Thanks for reporting. @jona-sassenhagen are you using `n_jobs > 1` ?

This is quite possible because it involves cross-validation. It will depend on how you do the splits. You can set the `random_state` like here: https://github.com/autoreject/autoreject/blob/master/examples/plot_visualize_bad_epochs.py#L96. That should hopefully fix it!

I didn't see your images by the way ... I think you intended to share something

hmm ... this is weird. I'm guessing you are using the latest code with bayesian optimization? If you can share your data privately, perhaps I can take a look.

If you want to be adventerous, you could [try changing](https://github.com/autoreject/autoreject/blob/master/autoreject/autoreject.py#L226) `n_calls` here to a higher number to see if it helps ... I haven't exposed this though and I'm not...

hmm ... I don't think so. Bayesian optimization should work better. It's just a bit slower but it should give more accurate solutions, especially if we increase the number of...

If you have any more questions, I will respond tomorrow. It's late in Europe :)

yes I would say so. Take a [look at this comment](https://github.com/autoreject/autoreject/issues/16#issuecomment-263053526). The method is a bit sensitive to how deep the cross-validation minima is. Your data looks pretty clean, that's...

What do you mean there is no way way to control verbosity? There is a `verbose` parameter, no?