RANSAC consumes too much RAM

[yjmantilla]

I have been testing pyprep against our datasets (GRUNECO, University of Antioquia, Colombia). For example I have 58 channels @ 1000Hz for about 5 minutes. This will need like 10 gb of RAM awarded from the OS to python. This is quite a lot for personal computers.

I have two possible proposals to solve this:

• Run RANSAC in a channel wise manner.
• Automatically identify the resample needed for the data to fit memory

I like more the first proposal and @sappelhoff agrees with this approach. This was discussed in #9 .

Once #9 detrend_correction is merged into master I will upload this code so it may be reviewed and tested. It considerably changes find_bad_by_ransac() and a tiny change in run_ransac()

[sappelhoff]

looking forward to a PR, I hope this would not slow down RANSAC too much 👍

[yjmantilla]

Im kind of academically compromised right now. I could do a pull request with the code as is, the problem is that is has no explanation. Or maybe by the time @adam2392 is available I will be more available too.

[yjmantilla]

Regarding #8 and #9 this is a comment to remind me that I forgot to include another do_detrend = False in another call to noisyChannels in the robust reference function. I will fix this in the channel-wise ransac PR.

Also include the seed for the ransac as the matlab one according to #21