Rémi Louf

Results 533 comments of Rémi Louf

I am sorry but I am not working on Aesara anymore. I am closing the PR for now.

You can try `t.type.shape` if the shape is known before evaluating the graph. The reason you don't get a tuple of int immediately is that in most cases the shape...

While the forward test passes **this shouldn't be merged**. For the backward transformation, many-to-one, I need to select a `rng`, `dtype` and `size` for the Half-Cauchy. I choose the inner...

More concerning to me is how to handle the `rng`s in these one-to-many relationship. I currently set the `rng` to both `InverseGammaRV` to the Half-Cauchy's, but that can't be correct....

> In this case, we only need consistency/reproducibility of the sampler's results between runs. Agreed. > I think the `kanren`/`etuple` aspects may only end up complicating things, It does, but...

The sum of normally-distributed random variables is partially implemented in this comment: https://github.com/aesara-devs/aemcmc/pull/66#issuecomment-1258471312

[ProbOnto](https://sites.google.com/site/probonto/) is a knowledge base that encodes many relations between probability distributions, which we should implement in our relational framework.

Yes. AeMCMC will also need the extra information that we'll add to `MeasurableVariable` eventually.

As far as I understand, you cast the _number of steps_ to `int64`. The high-level entry points of the library, and that you might want to use in a `Scan`...

For Gaussian scale mixtures, see for instance this paper: http://www2.stat.duke.edu/~mw/MWextrapubs/West1987a.pdf