WhatIsTheToolbox - PrincetonUniversity/princeton-mvpa-toolbox GitHub Wiki

What is the toolbox?

This is what you get when you hand over your money (price currently stable at $0).

Sample dataset

10 runs from a single subject of Haxby et al.'s (Science, 2001) 8-category study.

Note: we have made no efforts to check that the sample data set is valid or that it has been processed the same way as the published data. It's included for tutorial purposes only.

Tutorial and accompanying script

Sample script (tutorial_easy.m) that runs a basic analysis of the Haxby et al. dataset. This is accompanied by a line-by-line description of what's going on - this tutorial is the very best place to start with the toolbox.

Documentation

The docs subdirectory includes beautified html versions of the m-files, but this online wiki is really the best place to look.

Mathworks toolboxes

We optionally rely on the Mathworks toolboxes described below at some points in the toolbox. However, they are not bundled with the toolbox and must be purchased separately.

The non-parametric wavelet-based statistical functions (e.g. WAVESTRAPPER_RESULTS.M) require the Matlab Wavelets toolbox.

Statistics

By default, zscore_runs.m and statmap_anova.m require the anova1.m and zscore.m functions from the Statistics toolbox. However, we have provided home-grown alternatives that function similarly (though not identically). Just set the optional 'use_mvpa_ver' flags to true as described in the functions' help to use the home-grown functions instead.

Update: we've since realized that even our homegrown STATMAP_ANOVA.M relies on some of the scripts from the Statistics toolbox. If you have trouble with this, let us know. We're starting to recommend that people avoid the ANOVA and use the GLM from AFNI or otherwise anyway. See TutorialAfniGlm for info on how to do this. Also, if there any Octave experts out there, shout out and maybe we can see about borrowing some Octave code to function in place of the Statistics toolbox.

Neural networks

The Mathworks Neural Networks toolbox is required for the train_bp.m and test_bp.m functions, but you can use train_bp_netlab.m and test_bp_netlab.m instead, which rely on the bundled open source Netlab toolbox (see External bundled packages / Netlab below).

External bundled packages

We have been generously allowed to bundle the following packages by their authors. Although we have done some minimal testing, we haven't made any serious efforts to scrutinize their innards ' please let us know if you find a bug in them, but don't hold us responsible.

AFNI-Matlab

We rely heavily on Ziad Saad's AFNI-Matlab toolbox for all of the importing from and exporting to AFNI.

Netlab

If you don't have the Mathworks Neural Networks toolbox, you can use Ian Nabney's Netlab open source neural networks toolbox instead, although train_bp_netlab.m doesn't currently support all of the functionality in train_bp.m.

Boosting

By the time you read this, the wrapper for the boosting classifier should be ready (or available upon request). This requires the user to download the appropriate binaries. We're also working on a Matlab-only version.

Subversion

The Subversion (SVN) version control system is not bundled with the toolbox in any way, but we do rely heavily on it for internal development, and are extremely grateful for its existence.

Montage

We use Keith Schneider's montage library for visualizing brain maps. See Visualization Tutorial.