Rémi Louf

Just a note that #184 originally introduced a rewrite for `at.square` as well, but I think it would be better to only register `at.pow` here and make sure they're equivalent...

Yes. AeMCMC would transform the graph and specify the variables to condition on. But that might only make sense when AePPL is able to condition on transformed variables e.g. `at.exp(x_rv)`...

When it comes to probability distribution my vote goes to AePPL. Also `at.inv(weibull)` is nice but we should still have a way to give those a type so we can...

Would you say we should drop `LogNormal`, since it's just an exponentiated `Normal`? And `Binomial`? And `InvGamma`? The choice of what we name and what we leave as function of...

> I think the criteria will ultimately be: Does the specialized Op produce significantly faster/reliable draws (or logp graphs) than the derived one? I agree for the draws and the...

Implement a `Frechet` Op, say in AePPL, so that the following graph: ```python x_rv = aeppl.random.Frechet(1, 1, 1) ``` is equivalent to: ```python x_rv = at.inv(at.random.weibull(1, 1)) ``` as far...

So will we match `at.inv(weibull)` directly in our rewrites? And are we ok not having a `FrechetRV` object we can directly manipulate and represent in the graph? Then, what about...

It would be helpful if you could open an issue for each missing transformation!

Here are the missing transforms I identified looking at other PPLs. We should open individual issues for each: - [ ] Simplex constraint - [ ] Zero-sum constraint - [...

Afair _truncation_ means that you _cannot_ observe any data and that you know that before observation. _Censoring_ is when _we don't know_ the value of every data point, but only...