Though MALLET is the standard, I'm wondering if anyone has done testing on time to completion, based on document size, corpus size, varying hyper-parameters, topic count, etc.
I'd expect that different implementations would yield different qualities of output, something else worthy of comparison.
I'm also interested in runtime/output quality comparisons between basic LDA implementations and more complex topic modeling (e.g. Hierarchical LDA).
Thus far I've been comparing the basic C LDA from Blei's page against MALLET. With little surprise MALLET bests LDA C handily.
