Changelog

CoSMoMVPA changelog

• 1.1.0 (2016-12-06)

  • Fixed

    • Major Fix issue in cosmo montecarlo cluster stat, which returned z-scores corresponding than one-tailed instead of two-tailed probabilities, doubling the Type I error rate.

    • When using GNU Octave, let cosmo check external deal properly with packages that are installed but not loaded.

  • Changed behaviour

    • In cosmo crossvalidate, the first output contains (for each sample) a prediction for each fold, rather than one single prediction computed by cosmo winner indices. (Unchanged is that cosmo crossvalidation measure returns, by default, a single prediction per sample).

  • Deprecated

    • In cosmo crossvalidation measure, the “predictions” and “raw” output options are deprecated; use “winner_predictions” instead.

    • cosmo dataset slice fa and cosmo dataset slice fa will be removed in the future; use cosmo slice instead.

    • cosmo dim slice will be removed in the future; use cosmo slice and cosmo dim prune instead.

    • cosmo meta feature selection classifier will be removed in the future; use cosmo classify meta feature selection instead.

  • Removed

    • in cosmo crossvalidation measure, the “accuracy_by_chunk” output option has been removed. Use “fold_predictions” to get predictions for each fold.

    • in cosmo naive bayes classifier searchlight, the “predictions” output option has been removed. Use “winner_predictions” instead.

  • New features

    • cosmo meeg dataset and cosmo map2meeg provide input/output support for EEGLAB datasets with study structure, and for FieldTrip datasets with “subj” as the first dimension in dimord.

    • cosmo montecarlo cluster stat and cosmo searchlight support parallellization over multiple CPU cores/threads using the “nproc” parameter.

    • New function cosmo independent samples partitioner for MEEG datasets where samples can be assumed to be independent.

    • New function cosmo randperm with optional seed.

    • New functions for parallel computations on the Matlab and Octave platform, cosmo parallel get nproc available and cosmo parcellfun.

    • New functions cosmo map pca and cosmo pca for Principal Component Analysis (PCA), and support for PCA in cosmo crossvalidate and cosmo crossvalidation measure.

    • Added Changelog

  • Contributors

    • Tijl Grootswagers

  • Acknowledgements

    • Christoph Braun

    • Tijl Grootswagers

    • Thomas Hinault

    • Maria Montchal

    • Daria Proklova

    • Joscha Schmiedt

    • Jo Taylor

    • Lara Todorova

    • Raffaele Tucciarelli

• 1.0.0 (2016-07-22)

  • First public release.