A patched version of …/mev/cluster/algorithm/nonpar/Nonpar.java has been committed to the Trunk and the 4.1.01 branch in the SF SVN.
The Fisher Exact test in Nonpar wasn’t correctly handling evaluation of group assignments in cases where some of the loaded samples were ‘Excluded’ by the user. The subset of data was pulled from the matrix but downstream, during taking the tally for the contingency matrix, the original grouping information was used rather than the sample subset reflecting excluded samples. You can diff to the previous version to view the change.
This problem was found this morning and is confined to the Fisher’s Exact Test.
Software Engineer II
J. Craig Venter Institute