He Jia

> I just rewrote `evaluate`, `logpdf`, `pdf_marginal` and `logpdf_marginal`. Now all of them use `np.einsum` instead of looping over the data or points. I think it's not necessary for `evaluate`...

@rgommers @ilayn Now the problem is, without stuffs like tree-based computation, kde can be expensive for large data set. Previously it was looping over the data points or input points....

@mdhaber This is a quite old thread. It did not get merged because my naive implementation could not pass the tests due to memory and/or run time limits. I no...

> Thanks again @HerculesJack. One last question: this PR adds methods for computing the pdf/logpdf of a marginal distribution of a KDE distribution. I'm thinking that rather than having a...

Sorry for the delayed reply. In the example above making `spin` static in jit is not a good idea for me, since I do need this to work at many...

@mblondel not sure if I understand your comment regarding `stop_gradient`. Nothing changes if I do `lower=jax.lax.stop_gradient(r_0), upper=jax.lax.stop_gradient(r_1)` in my snippet.

Sorry for the documentation issue! We'll probably have a better doc in a few weeks. For now, I suppose you can do it in this way: ``` import bayesfast as...

This is related to the issue that sometimes it gets logq sample at which logp is not defined. The DES example is not working after I upgrade to python 3.8...