What steps will reproduce the problem?
1. Running gaussFit as a subroutine and using a prior (for me through
hmmFitFullyObs).
What is the expected output? What do you see instead?
The updating of the gaussian-InverseWishart distribution appears to be
incorrect and yields a highly non PSD result.
What version / revision of the product are you using? On what operating
system?
pmtk3 - Nov2012/Ubuntu
Please provide any additional information below.
Replaced line 37: SN = S0 + C + kappa0*m0*m0'-kappaN*muN*muN';
with
SN = S0 + C + kappa0/(kappa0+N)*(xbar-m0)*(xbar-m0)';
The problem appears to be fixed
Original issue reported on code.google.com by miketrav...@gmail.com on 23 Aug 2013 at 8:33
Original issue reported on code.google.com by
miketrav...@gmail.com
on 23 Aug 2013 at 8:33