Improve testing for resampling in the autocorrelation feature

[mscheltienne]

It looks like I got it right in #21.
WDYT?

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (main@24e190f). Click here to learn what that means.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main      #26   +/-   ##
=======================================
  Coverage        ?   97.30%           
=======================================
  Files           ?        7           
  Lines           ?      408           
  Branches        ?        0           
=======================================
  Hits            ?      397           
  Misses          ?       11           
  Partials        ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 24e190f...9a81494. Read the comment docs.

[adam2392]

Wouldn't this only handle sampling rates of 100 Hz? What if it was 200?

Maybe we should try handling all common sampling rates of EEG hardware? So I'm guessing like 50, 100, 128, 200, 256?

Idk, can get complicated though :/

[mscheltienne]

Well, that's what EEGLAB does. The autocorrelation functions ends with:

% resample to 1 second at 100 samples/sec
resamp = resample(ac', 100, EEG.srate)';

Which corresponds to resample_poly with up hard coded to 100, and down set to sfreq.

All it's doing in #21 and in this PR is supporting floating point sfreq by setting the down argument either to int(sfreq) or to int(sfreq) + 1, in order to get the correct output shape (n_comp, 101).
In EEGLAB, since the sfreq is always an integer, this step is not needed because if down=EEG.srate (int), then the output shape is always (n_comp, 101).

[adam2392]

Well, that's what EEGLAB does. The autocorrelation functions ends with:

% resample to 1 second at 100 samples/sec
resamp = resample(ac', 100, EEG.srate)';

Which corresponds to resample_poly with up hard coded to 100, and down set to sfreq.

All it's doing in #21 and in this PR is supporting floating point sfreq by setting the down argument either to int(sfreq) or to int(sfreq) + 1, in order to get the correct output shape (n_comp, 101). In EEGLAB, since the sfreq is always an integer, this step is not needed because if down=EEG.srate (int), then the output shape is always (n_comp, 101).

Fair I guess that's what ICLabel does so at least for this version! Unfortunate, I wonder how sensitive ICLabel is to the sampling rate... Room for improvement :)