this doesn't seem very well documented at present.

Summary:

Regarding the parameters of the dual averaging optimization for the stepsize in the warmup, all parameters can be set by the user except mu. For certain models, the initial choice of mu can result in a drastic drop of the stepsize for subsequent iterations. This might lead to extra computation time. As mu is adjusted for each window of the mass matrix adjustment, this can in e

Ankit Shah and I are trying to use Gen to support a project and would love the addition of a dirichlet distribution

This looks really awesome and would love to get started with this. Is there an official docs page?

the links in the README pull up HTML pages that seem to be missing images for the models.

Perhaps I could convert them to markdown/latex cells so they'll guarantee rendering?

Love this repo, by the way. Writing a software library (and demos for it) based on my thesis work in inverse problems, and this helps motivate what "good tutorials" can look like.

@bgoodri I just noticed that none of the CRAN versions of the vignettes have any output from the code chunk no output anymore (e.g. if you look at https://cran.r-project.org/web/packages/rstanarm/vignettes/count.html
or any of the other vignettes). That is, it looks like none of the code is getting evaluated. I know that we're using that mechanism with params$EVAL to avoid evaluating the vignet

To do.

Just noticed that this isn’t discussed anywhere, but the PPC functions have always also been very useful for prior predictive checking, not just posterior checking.

pd documentation states that

It varies between 50% and 100% (i.e., 0.5 and 1)

However, when exactly 0 has high credibility, it is actually possible to go below 50%.
This can happen when using model averaged posteriors (out weighted_posteriors or brms::posterior_average).

I think this case should be addressed in the function docs.

library(bayestestR)
library(rstanarm)

For example,
in ELFI tutorial...
We should specify observed data like below:

Y = elfi.Simulator(MA2, t1, t2, observed=y_obs)

for calculating a discrepancy like:

d = elfi.Distance('euclidean', S1, S2)

What I wonder is,
without explicitly passing 'observed' argument to the Simulator object,
can we make it possible to do ABC-rejection by modifying discrepancy function?
e.g. inste

Hi,

Thanks for sharing the code.

Is there ant simple "getting started" document on setting up like a basic EI or KG with your tool.
1- I have seen the main.py in the "example" folder, it is good but it is for a very complete case, not for simple sampling mode.
2- I have tried the MOE getting started document (available in the "doc" folder) but as the function names and procedures are chan

There was recently some discussion on Stan Discourse that resulted in this PR in Stan and this PR in PyMC3. They make some changes to the NUTS criterion to handle a previously undiscovered case where the N

From an older issue
@infotroph about documentation: https://www.divio.com/blog/documentation/

More context: This piece makes a strongly-argued case that there are four distinct types of software documentation, and that all well-documented projects need to have all four of them as explicitly separate sections:

• tutorials, for teaching beginners what your tool does using step-by-step example