Future #13
Future #13
Conversation
Codecov Report
@@ Coverage Diff @@
## master #13 +/- ##
==========================================
+ Coverage 98.12% 99.09% +0.96%
==========================================
Files 5 3 -2
Lines 213 110 -103
==========================================
- Hits 209 109 -100
+ Misses 4 1 -3
Continue to review full report at Codecov.
|
…iscrete parameters
* docs * examples * push back particles onto domain.
|
@jbrea what do you think about this? |
There was a problem hiding this comment.
Cool, nice work!
So far I went only very quickly over this. I like the idea of making it more general based on the interface functions accept and loglike. But I would need to think a bit more about the definitions of the structures, e.g. putting costs into ApproximatePosteriors didn't seem too intuitive to me at first glance and maybe it would make sense to define some structs for the particles?
Also, shall we look into AbstractMCMC.jl as you mentioned and was brought up again in JuliaApproxInference/ApproxInferenceBase.jl#1?
Please let me know if you want me to look in more details at this PR or something in particular.
| densitytypes(S::DiscreteDistribution) = Int64 | ||
|
|
||
| floatize(S) = floatize.(S) | ||
| floatize(S::Number) = convert(Float64, S) |
|
I agree on both points, a in this case, i will merge this, and slowly work on another PR to include a |
The aim of this PR is completely refactoring all algorithms into a single powerful algorithm with multiple available moves,
inspired by Differential Evolution MCMC, Affine Invariant MCMC.
whether this will be merged in full, in part, or not is still to be decided.