Tiem van der Deure

Has there been any progress on this? Multiplying a Cholesky matrix with a Diagonal does not work with ReverseDiff, e.g. in the example shared here: https://github.com/TuringLang/Turing.jl/issues/1870#issuecomment-1706595909

Sorry for being imprecise. The `LowerTriangular` comes from a `Cholesky`, though. It was recently implemented (which is great!) and I suspect I won't be the last one to try a...

I'll try with `Bijectors.pd_from_lower`, thanks! > Also, just for the record, both of the examples you give fail on my end (which is what I expected) 😕 I made a...

Using `Bijectors.pd_from_lower` I quickly ran into numerical stability problems, especially with bigger covariance matrices. E.g. ``` @model function model_with_filldist(i) stds ~ filldist(truncated(Normal(0, 0.5); lower = 0.0), i) F ~ LKJCholesky(i,...

I'm still getting some weird behaviour when using `LKJCholesky` with compiled ReverseDiff. `benchmark_model` doesn't identify any gradient differences, and sampling is much more inefficient in compiled mode `NUTS`. I know...

Any news on a fix for this issue?

Hi, thanks for chipping in. I don't think the regularization lambdas is where the differences are coming from. In the maxnet algorithm the lambdas are generated inside the algorithm and...

I think the test failures are unrelated to this PR?

Thanks for reviewing. I'll add some tests and documentation later this week!

I just added some quick docs - most of it is copied over from the docs for `NeuralNetworkClassifier`. Should I still wait for https://github.com/FluxML/MLJFlux.jl/pull/251 before I can update the tests?