Hi, On Thu, Jun 30, 2011 at 2:26 PM, Jim Kleckner <[email protected]> wrote: > Professor Stein, > > I downloaded Sage recently and have been very impressed during my > explorations! > > In digging into the hmm code, in particular GaussianMixtureHiddenMarkovModel > class, I noted a couple of trivial documentation fixes for which a patch > file > is attached.
Can you get an account on trac.sagemath.org, make a ticket, and put the patch there? This is the standard way hundreds of people do Sage development, and will make it easier when you want to submit things in the future. Let me know if you do this, so I can referee your patch. How to get an account is explained at http://trac.sagemath.org/sage_trac/ at the top. > I had some questions about the implementation and if you are not interested, > just > let me know and I won't bother you. > Also, tell me if I should take this to a > discussion list and if so which one is best. I'm cc'ing this message to the sage-devel list, which is where such discussion belongs. > I was interested in using the code for a problem that has vector-valued > rather > than scalar observations. The reference work that I'm reading that spells > out > the technique is: "Markov Models for Pattern Recognition", by Gernot A. Fink > > The formulas are detailed on page 85 here: > http://books.google.com/books?id=L4EgpAl2GWgC&lpg=PP1&dq=gernot%20fink&pg=PA85#v=onepage&q&f=false > > Do you know of existing implementations that work with vector-valued > observations > that are out there? I personally don't. > Can you think of a way of using the existing code by some sort of > composition to > accomplish > > How much work would it be to extend your implementation handle vector > observations? I don't know, but I strongly encourage you to try. > > It appears that what is needed is: > * Extend GaussianMixtureDistribution to become, say > GaussianVectorMixtureDistribution, which supports the dimension of the > observation vector with a vector of (mu, sigma) and covariance matrix > among them. > * Compute the covariance matrix Cjk per the equations in Fink. > * Figure out how to implement prob() and sample() for the vector > distribution implementation. > * Decide whether to continue to use TimeSeries (for speed) or > revert to numpy for ease of implementation. > * ?? > > Does this seem correct? I don't know. > Thank in advance for your advice. Sorry that I don't really have any advice. You are probably the third person to personally ask me about having vector valued GHMM support in Sage, so it's something a lot of people want. But I've never thought at all about how to implement it, hence don't have anything interesting to say. -- William Stein Professor of Mathematics University of Washington http://wstein.org -- To post to this group, send an email to [email protected] To unsubscribe from this group, send an email to [email protected] For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
