PipeCompare 0.9.4

• The results say whether the recommended pipeline is practically as good as the best one (its
  noise within 5 %) or whether there are too few trials to tell, and how much noisier it could
  be (equivalent, diffHiKept in the ranking table; equivalenceMargin option).
• Selection check: the best pipeline is also scored on trials that were not used to choose it
  (cross-fitting over 20 random half splits, nSplits option), and the results say how much
  better it looks than it is (ranking.byStratum.crossfit).
• Rejection limit chosen from the data (reject_threshold uv = auto): every limit that keeps
  a different set of epochs is scored and the exact best is found; the highest limit the data do
  not distinguish from it is used. The simple mode uses it by default (36 pipelines for
  Standard instead of 108, 3 for band power instead of 9); with repaired epochs or peak
  measures, fixed limits are still compared.
• ICLabel threshold chosen from the data (icremove threshold = auto): every threshold that
  removes a different set of components (all components over 0.5) is scored together with the
  best rejection limit for it, the gain and signal check coming from the known signal with the
  same components removed; the exact joint best is found, and the threshold removing the fewest
  components that the data do not distinguish from it is used. The simple mode uses it by
  default (12 pipelines for Standard instead of 36, 1 for band power instead of 3); with
  repaired epochs or peak measures, the thresholds 0.7, 0.8 and 0.9 are still compared. When it
  removes nothing although ICLabel took some components for artifacts, the results say so.
• Filter check: what each pipeline's filters alone do to the known signal is computed before
  anything runs, on a short stretch of the signal copy; pipelines whose filters alone break a
  signal-check limit are excluded without being run (reason filters alone: …; option
  filterCheck).
• Simple mode: the data description at the top has a fixed height and scrolls. A long montage
  check used to squeeze the event-type list to a sliver, so no event type could be chosen and
  Run stayed grey.
• The progress window is one plain window with its text, bar and Stop, instead of MATLAB's
  progress box drawn inside a smaller window (which cut off its bottom edge).
• The result window of the simple mode: the recommendation at the top has a fixed height and
  scrolls, so its notes can no longer squeeze the table of pipelines; the window is a little taller.