-
Notifications
You must be signed in to change notification settings - Fork 121
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Problem with solution of maximum entropy problem (or request for documentation) #64
Comments
Thanks for taking the time to write some examples! If you run SCS with
So, we're hitting the maximum number of allowed iterations. If you instead use, say I don't think |
@karanveerm, usually the |
@gajomi, we're starting to collect example notebooks at https://github.com/JuliaOpt/juliaopt-notebooks, if you'd like to put together this example with a bit of explanation and LaTeX math, the contribution would be more than welcome. |
@karanveerm - Thanks. This fixes my problem. @mlubin - Yeah, I can put together a little notebook about all this. I should have some time next week. |
I thought it would be nice to contribute a pair of examples for finding exponential family distributions (http://en.wikipedia.org/wiki/Exponential_family) with a given set of features and expectation values. The first example would consist of maximizing the relative entropy between the target distribution and a base measure under constraints n the expectation values. The second example would be the same thing, but where the maximization occurs on a much lower dimensional dual space (the Langrange multipliers that fix expectation values for a set of chosen features), which presumably is the easier problem. The point of the example would be to show that the two methods yield the same result.
Here is a bit of code that (at least initially), appeared to solve the first part of the problem for the case where the solution should be a Binomial distribution
However, trying out a larger number of states (N=64) results in:
In both cases problem.solution.status is :Optimal. However, from the figure above it is clear that the solution is considerably far from the linear constraint. So, is there a bug (in convergence or constraint enforcement) at larger N values or do I just have the wrong idea about the range of final states that yield :Optimal convergence status? If the later is the case, is there some documentation one sensitivity parameters to ensure one is near the constraint subspace?
P.S.
This is my first time looking into Convex.jl and this whole thing about automatic convexity detection within such a rich modeling language seems very magical. Its great :)
The text was updated successfully, but these errors were encountered: