Added cholesky to MVGaussianRandomWalk and MVTRandomWalk

[sachinruk]

As requested by #2569 creating a pull request to include cholesky decomposition of the Covariance matrices.

[twiecki]

This looks good. Thanks for the contribution @sachinruk. Can you add a test?

[sachinruk]

@twiecki could you show me an example test that was already there for normal MvGaussianRandomWalk? I can base my test off of that. Sorry my coding skills aren't the best.

[twiecki]

Hm, I guess there is no existing unittest. Those are probably also hard to write in this case. Alternatively, can you post an example NB that shows this works?

[sachinruk]

Done! Hope it's all good. https://github.com/sachinruk/pymc3/blob/chol_MV/docs/source/notebooks/MvGaussianRandomWalk_demo.ipynb if you want to see the Notebook before you accept.

[junpenglao]

LGTM, @aseyboldt any comments?

[junpenglao]

Is there any reason you are not going with the solution you proposed in #2569 using class MvGaussianRandomWalk(_QuadFormBase)?

[sachinruk]

From memory, I did try that but got an error which I forget what it is. _CovSet is a subset of the functions in _QuadFormBase. Actually I don't think I could have imported _QuadFormBase, because it was a hidden class. _CovSet is a subset of _QuadFormBase where just the functions that are necessary for MvGaussianRandomWalk have been copied across.

[sachinruk]

Hey peeps, sorry to bug you all but is there anything more to do here?

[junpenglao]

Forgot about this - thanks for the contribution!