Add epochs.regress_evoked function

[wmvanvliet]

Remove ERF from single trial using ordinary least-squares regression i.e., minimizes
the sum of squared differences between the predicted and observed trial
signal.

[1] Fit ERP waveforms to single trial
[2] Use β coefficient, estimate the presence of ERP in the data
[3] Subtract only the β ERP from the single trial (not the full average)

        epochi​(t)=βi​⋅ERP(t)+εi​(t)

To minimize contamination of induced spectral estimates by phase-locked
activity, we applied regression-based ERP removal to each trial prior to
spectral decomposition. Unlike the subtraction of the averaged evoked
response implemented in mne-ython
(https://mne.tools/stable/generated/mne.Epochs.html#mne.Epochs.subtract_evoked)
, this regression models the trial-specific contribution of the ERP, which
preserves the non-phase-locked (induced) oscillatory activity while
removing the shared evoked component (e.g. Grandchamp & Delorme, 2011; Cohen,
2014). This approach is particularly important when computing 1/f and
oscillatory components using IRASA, as residual ERP may bias the estimation
of aperiodic activity and oscillatory peaks (Makeig et al., 2004).

• Grandchamp, R., & Delorme, A. (2011). Single-trial normalization for event-related
  spectral decomposition reduces sensitivity to noisy trials.
  Frontiers in psychology, 2, 236.

• Makeig, S., Debener, S., Onton, J., & Delorme, A. (2004). Mining event-related
  brain dynamics. Trends in cognitive sciences, 8(5), 204-210.

• Cohen, M. X. (2014). Analyzing neural time series data:
  theory and practice. MIT press.

[wmvanvliet]

@virvw

[wmvanvliet]

We're holding off on this for now until it is used in a publication.