AbstractMCMC.jl

AbstractMCMC.jl copied to clipboard

Improving interaction between different implementations of the interface

7

How do you feel about adding something like: ```julia """ state_from_transiton(state, transition_prev[, state_prev]) Return new instance of `state` using information from `transition_prev` and, optionally, `state_prev`. Defaults to `setparameters!!(state, parameters(transition_prev))`. """...

torfjelde

Why `step` initializes the chain?

6

Hi, Under the current API, one implements two specializations of `step`: one for initializing the chain, and one for actually "stepping". But why didn't we make a separate interface for...

Red-Portal

Addition of `step_warmup`

14

For many samplers, it might be useful to separate between the warmup phase and the sampling phase, e.g. in AdvancedHMC we have an initial phase where we adapt the parameters...

torfjelde

Relating to After https://github.com/TuringLang/AbstractMCMC.jl/pull/119 we were discussing whether we wanted to rely on FillArrays.jl like this or not, but decided to do so for now. Opening an issue to keep...

torfjelde

Add `state_from_transition`, `parameters` and `setparameters!!`

17

See #85. Currently, almost every package implementing a `AbstractSampler` comes with their custom state and transition types. Since there currently are no "interface" or "guidelines" about what the types of...

torfjelde

I need to sample a distribution for which the calculation of the log-likelihood is particularly time consuming. To speed-up calculations, it is beneficial to pass to the likelihood mutable data...

astrozot

Make `AbstractMCMC.step` function handle `rng` as part of state

5

For continuing MCMC sampling from a previous stopping point, we need to store the `rng` as part of the sampling state. https://github.com/TuringLang/AdvancedMH.jl/blob/e1741179e2505da57945d47b7b1debbf3f0e848b/src/mh-core.jl#L83 https://github.com/TuringLang/AdvancedMH.jl/blob/e1741179e2505da57945d47b7b1debbf3f0e848b/src/mh-core.jl#L90

yebai

Correct syntax for chainscat and chainsstack

3

I'm not sure what's the correct syntax for the functions `chainscat` and `chainsstack`. I tried calling `chainscat(a,b)` and `chainsstack([a,b])` and both return an error. (As an aside, is the only...

astro-kevin

Default `bundle_samples` is quite annoying

1

The default implementation of `bundle_samples` specializes on `Vector{T}`, which in becomes very annoying if you want to implement your own `bundle_samples`, e.g. if you then use `AbstractVector` you almost immediately...

torfjelde

Consider transfering the `ESSModel` to `AbstractMCMC`

2

We currently have a type [`ESSModel`](https://github.com/TuringLang/EllipticalSliceSampling.jl/blob/a76718df4894c54d84711b2e45a310167a01b7c1/src/model.jl#L3 ) that implements a factorised version of `logdensityproblem`. Now that we have the following new type, does it make sense to generalise and transfer...

yebai