You are viewing a plain text version of this content. The canonical link for it is here.

Posted to user@mahout.apache.org by fraser wilfred <fr...@gmail.com> on 2013/03/26 15:20:31 UTC

LDA perplexity measure - what is a good value?

I've been running the perplexity checks on LDA for my data, but I'm
slightly confused as to what a good or normal perplexity range is. In
understand that the lower the score the better. Lots of papers I've looked
at have perplexity values in the range of thousands, but my data is
producing values in the hundreds, example output below:

 CVB0Driver - Read 1 entries with total perplexity 17312.95783869685 and
model weight 131.99999999999991
CVB0Driver - Current perplexity = 131.15877150527925

Is this within the normal range for perplexity scores?

Many Thanks,
Fraser