nsh87 / receptormarker

Source for 'receptormarker' package for R: antibody receptor and phenotypic marker analysis
http://receptormarker.com
BSD 2-Clause "Simplified" License
4 stars 9 forks source link

Use NbClust to determine optimal k and update multiClust structure to an S4 class #66

Closed nsh87 closed 8 years ago

nsh87 commented 8 years ago

WSS Average silhouette score was being used to determine the optimal k when running multi_clust(). This updates NbClust's code so that it is more reliable when running complex or simple data sets through it and switches out the determination of optimal k to use NbClust, instead. NbClust runs 20+ algorithms and each one nominates an optimal k. The final optimal k is determined by majority rule of the algorithms.

This also updates the multiClust structure to an S4 class and adds documentation for the class. Previously, multiClust was not a class and therefore not documented, and it could be overruled by a variable of the same name in the user's scope.

Do not merge until:

nsh87 commented 8 years ago

@catterbu: it has been a while, so just want to remind you of this PR. i was looking for where all the NbClust stuff was, it's here. might be worth taking look at the issue in the PR description, too.

nsh87 commented 8 years ago

@catterbu: screw it, i'm going to merge this then try to apply your fix from last night to not use NbClust.

Note: this does NOT resolve Issue #67. We need to add an option to the function method='fast' or method='exhaustive' to use silhouette score or NbClust.