[mlpack-git] [mlpack/mlpack] How to organize the data? (#809)

[Ryan Curtin]

Hi there,

If you are using a discrete distribution (`--type discrete`), then multidimensional observations aren't supported.  It would not be impossible to support multidimensional discrete distributions (but there will be no relation between dimensions), but currently the code doesn't do that.  I think that we should leave this issue open until multidimensional support is added for HMMs.  For now, you might consider "collapsing" your dimensions: suppose that each of your three dimensions takes values between 0 and 7 (so, 3 bits).  Then given `(a, b, c)` you map to a single-dimension value of `2^6 * a + 2^3 * b + c`.  That will cost performance but it will be a workaround for now.  You could also use a Gaussian HMM if the dimensions are not categorical (`--type gaussian`).

Sorry that the support you are looking for is currently not there.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/mlpack/mlpack/issues/809#issuecomment-258858129
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.cc.gatech.edu/pipermail/mlpack-git/attachments/20161107/8836b6fe/attachment.html>