Xianda Sun
Xianda Sun
Thanks a lot, @torfjelde very clear!
the test is failing because an numerical test (https://github.com/TuringLang/AdvancedMH.jl/actions/runs/11434545096/job/31870916921?pr=103#step:5:353) should we relax the `atol`, and is this new?
I am not certain here: these functions are not exported, but it's true that through `AbstractMCMC` they become a part of the public interface (but in a softer sense). Not...
good call! now removed, thanks a lot for reviewing this all these times
CI errors look like not our faults. I'll take a deeper look shortly, thanks!
I buy @torfjelde's reasoning here. I also agree that the former of the two conditioning syntax is better. For one, `xs[1]` doesn't appear in the program. Also, I think in...
a bump on this
let me take a look, I haven't deal with this part of the code before, so give me bit of time.
The following version passes the test ```julia function AbstractMCMC.step( rng::AbstractRNG, model::AbstractMCMC.LogDensityModel, spl::SGHMC, state::SGHMCState; n_adapts::Int=0, kwargs..., ) if haskey(kwargs, :nadapts) throw( ArgumentError( "keyword argument `nadapts` is unsupported. Please use `n_adapts` to...
Looks fine to me. Now the tests pass, I think the algorithm is very likely to be correctly implemented. Still curious why [this](https://github.com/TuringLang/Turing.jl/blob/f184d3f4315a3808761485357794a8260fe8d1b7/src/mcmc/sghmc.jl#L95-L99) works (I am referring to the order...