pmtk3
pmtk3 copied to clipboard
Adaptive Rejection Sampling incorrect
From [email protected] on August 27, 2013 19:20:23
Hi,
I found a bug in the adaptive rejection sampling code -- ars.m
arsEvalHulls returns log-values but they are subsequently treated as probability values for the accept/reject steps in lines 86 and 91 (eg divided with each other and compared to a uniform draw). This can cause the ARS to sample heavier tails than it should.
Original issue: http://code.google.com/p/pmtk3/issues/detail?id=61