Infolds vs outfolds in CosmoMVPA


I’m running CosmoMVPA with an unbalanced set (say, 25 men and 75 women) with nmin=1.

This means that the partitions field has two layers:
1- infolds - folds that contain 1 run for test and all others for train (say 6 runs so 6 infolds)
2- outfolds - for each infold, there will be 3 outfolds with 25 men and 25 women.

My questions is the how are the outfolds treated -

  • are they concatenated to a single (in)fold (thus 6 folds in total)?
  • are the chained to get 18 folds in total but with less training samples in each fold?
  • is it something else?