Closed junotk closed 11 years ago
Fixed in develop
branch.
But cluster algorithms use metric, and seemingly convert similarity to dissimilarity --- very possibly with 1-x. That must be checked as well.
We will use Jansen-Shannon divergence for the cluster algorithms, so there will be no problem.
An S675 student pointed me out that to convert cosine similarity to dissimilarity one must use arccos(x), rather than 1-x. We are using dissimilarity matrices in isomap, so they need to be fixed.
http://inpho.cogs.indiana.edu/datablog/the-shape-of-philosophy-pt-1/#comment-265