Add helper function for easier Random Walks specification

[AlexAndorra]

This issue is up for grabs 🥳

Currently, users can use theano.tensor.cumsum() to create any random walk they need. We do have pm.GaussianRandomWalk though, which is a user-friendly wrapper but mostly allows one to specify a different init distribution, which is possible but awkward with the .cumsum() approach (it requires tt.stack()).

A nice addition to the library would be to turn this into a small helper function, like pm.RandomWalk('x', pm.StudentT.dist(), init=pm.Flat.dist()). This would give more adaptability to the way users can specify random walks 🎰🚶‍♂️

Feel free to signal your interest here or ask any questions if you want to work on a PR for this 🖖

[chandan5362]

Hi @AlexAndorra , is this issue still open? If it is so,
I would like to work on it and a little more information about the issue would be deeply appreciated.

[AlexAndorra]

Hi @chandan5362, yes this is still up for grabs, and thanks for your interest!

However, let me first check with @brandonwillard and @michaelosthege that it's still worth the investment: the coming v4 will use the new RandomVariable op, that all distributions will use instead of the current pm.Distribution.
So, guys, is it worth doing what's in the issue description and add this new wrapper, or will it have to be completely revamped anyway when switching to the Aesara backend?

[brandonwillard]

I would definitely recommend that efforts be directed toward Aesara and the PyMC3 RandomVariable updates. After a certain point, those changes will radically simplify timeseries work.

[chandan5362]

I am looking forward for your response @AlexAndorra .

[AlexAndorra]

I'd defer to Brandon's opinion then, what he's saying makes sense -- @twiecki may have insights on this too.
If we put a pin on that issue while v4 comes, we can redirect you to issues converting pm.Distribution to the new RandomVariable when they appear @chandan5362. This will be the backbone of the new version and, as Brandon said, will simplify the timeseries work underlying this current issue!

[chandan5362]

well, thanks @AlexAndorra
I will be happy to work on such issue. Please do ping me when it appears.

[ricardoV94]

Done in #6072 and #6131