When I tried to use SinghalSampler to calculate eigenvalue errors for MSMs, I got values that were not correct (they didn't match Dan Ensign's ssaCalculator results).
Nina assumes that the U matrix has unit diagonal, but scipy computes it such that L has the unit diagonal. I think that this will give a different result, but I suppose I could be wrong. Anyway, that may or may not be the cause of the errors.
rmcgibbo
commented
11 years ago
When I tried to use SinghalSampler to calculate eigenvalue errors for MSMs, I got values that were not correct (they didn't match Dan Ensign's ssaCalculator results).
I think I know why. I have since rewrote it just following Nina's paper (see https://github.com/schwancr/msmbuilder/blob/msm_error/src/python/error/singhal.py). I think the problem was when you computed the LU decomposition.
Nina assumes that the U matrix has unit diagonal, but scipy computes it such that L has the unit diagonal. I think that this will give a different result, but I suppose I could be wrong. Anyway, that may or may not be the cause of the errors.