search

tpapp / DynamicHMC.jl

Implementation of robust dynamic Hamiltonian Monte Carlo methods (NUTS) in Julia.
Other
244 stars 21 forks source link

Registrator release #46

Closed andreasnoack closed 5 years ago

andreasnoack commented 5 years ago

The current version has capped TransformVariables at 0.2 because of how the conversion script from METADATA works so it would be great with a new release that allows for TransformVariables >= 0.3

StefanKarpinski commented 5 years ago

I can make a PR like https://github.com/JuliaRegistries/General/pull/256 to fix this. What should I do? Change the upper bound on all packages that depend on DynamicHMC from 0.2 to 0.*?

tpapp commented 5 years ago

I think rather for all packages that depend on TransformVariables, to 0.*.

StefanKarpinski commented 5 years ago

So I see the following dependencies on TransformVariables:

D/DiffEqBayes/Compat.toml
45:TransformVariables = "0-0.2"

D/DynamicHMC/Compat.toml
19:TransformVariables = "0-0.2"

D/DynamicHMCModels/Compat.toml
22:TransformVariables = "0-0.2"

L/LogDensityProblems/Compat.toml
["0-0.8.0"]
9:TransformVariables = "0-0.2"
--
21-["0.8.1"]
22:TransformVariables = "0.2"
--
24-["0.8.2-0"]
25:TransformVariables = "0.2-0"

S/StatisticalRethinking/Compat.toml
47:TransformVariables = "0-0.2"

Presumably all the 0-0.2 ones should be changed to 0 and the 0.2-0 one should be left alone. What about the 0.2 compat entry? Should that be changed to 0.2-0 or left alone? In other words, do you want LogDensityProblems 0.8.1 capped at 0.2 or uncapped and free to use 0.3 and higher?

tpapp commented 5 years ago

Sorry, I don't understand what 0.2-0 means in this context.

But it is very likely that 0.3 is a drop-in replacement for 0.2 for most people, so anyone who used 0.2 should be fine with 0.3.

That said, to be conservative, we can just fix it for my own packages, DynamicHMC and LogDensityProblems. There, 0.* would be ideal.

StefanKarpinski commented 5 years ago

0.2-0 means that LogDensityProblems is compatible with an TransformVariables version in [0.2.0, 1.0.0).

StefanKarpinski commented 5 years ago

PR created: https://github.com/JuliaRegistries/General/pull/265.

StefanKarpinski commented 5 years ago

I didn't change anything for LogDensityProblems because it was unclear but that's also your package so it seems like you could say whether the bounds should be adjusted or not there.

tpapp commented 5 years ago

Closed by https://github.com/JuliaRegistries/General/pull/265/