Fix discrete init

[franknoe]

• bugfix: ensuring that the discrete HMM initialization provides a reversible transition matrix
• Added test using an initialization of a 3-state HMM
• bugfix: avoiding infinite recursion in desc and str of HMM subtypes due to inheritance from output model

[jchodera]

Looks good, and seems to pass! Commit when ready.

I bet we can cook up a test model with exact reference properties at some point, but this test seems fine for now.

[franknoe]

In this particular way of initializing the discrete HMM the problem is that we use PCCA to construct a initial transition matrix, but that is not guaranteed to give a nonnegative matrix, so I just modify it a posteriori using a simple heuristic to force it to be a transition matrix. So there's not really a meaningful exact reference unless one modifies the method.

Am 20/05/15 um 17:16 schrieb John Chodera:

Looks good, and seems to pass! Commit when ready.

I bet we can cook up a test model with exact reference properties at some point, but this test seems fine for now.

— Reply to this email directly or view it on GitHub https://github.com/bhmm/bhmm/pull/41#issuecomment-103922832.

---

Prof. Dr. Frank Noe Head of Computational Molecular Biology group Freie Universitaet Berlin

Phone: (+49) (0)30 838 75354 Web: research.franknoe.de

Mail: Arnimallee 6, 14195 Berlin, Germany