Closed
Conversation
This reverts commit 2020f2d.
Contributor
|
AdvancedHMC.jl documentation for PR #481 is available at: |
… gradients, but not changing the schedule). (#473) * initial changes to get a working demo * fix tests, add new ones, and add documentation * fix type * fix some stray tests * delete tmp folder with demo * address review comments * reference interface refactor issue * refactor and fix test rng handling * improve docstring for NutpieVar * remove superfluous white space * fix JET tests * add entry to history, bump version * fix NutpieVar docstring * increase number of tests for mass matrix adaptation --------- Co-authored-by: Markus Hauru <markus@mhauru.org>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Comment on lines
+6
to
+8
| currently to be initialized for a `metric` of type `DiagEuclideanMetric` | ||
| via `mma = AdvancedHMC.NutpieVar(size(metric); var=copy(metric.M⁻¹))` | ||
| until a new interface is introduced in an upcoming breaking release to specify the method of adaptation. |
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
| currently to be initialized for a `metric` of type `DiagEuclideanMetric` | |
| via `mma = AdvancedHMC.NutpieVar(size(metric); var=copy(metric.M⁻¹))` | |
| until a new interface is introduced in an upcoming breaking release to specify the method of adaptation. | |
| currently to be initialized for a `metric` of type `DiagEuclideanMetric` | |
| via `mma = AdvancedHMC.NutpieVar(size(metric); var=copy(metric.M⁻¹))` | |
| until a new interface is introduced in an upcoming breaking release to specify the method of adaptation. |
|
|
||
| - Adapt the mass matrix `metric` of the Hamiltonian dynamics: `mma = MassMatrixAdaptor(metric)` | ||
|
|
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
Comment on lines
+38
to
+40
| currently to be initialized for a `metric` of type `DiagEuclideanMetric` | ||
| via `mma = AdvancedHMC.NutpieVar(size(metric); var=copy(metric.M⁻¹))` | ||
| until a new interface is introduced in an upcoming breaking release to specify the method of adaptation. |
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
| currently to be initialized for a `metric` of type `DiagEuclideanMetric` | |
| via `mma = AdvancedHMC.NutpieVar(size(metric); var=copy(metric.M⁻¹))` | |
| until a new interface is introduced in an upcoming breaking release to specify the method of adaptation. | |
| currently to be initialized for a `metric` of type `DiagEuclideanMetric` | |
| via `mma = AdvancedHMC.NutpieVar(size(metric); var=copy(metric.M⁻¹))` | |
| until a new interface is introduced in an upcoming breaking release to specify the method of adaptation. |
|
|
||
| - Adapt the step size of the leapfrog integrator `integrator`: `ssa = StepSizeAdaptor(δ, integrator)` | ||
|
|
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
|
|
||
| - The randomness is controlled by `rng`. | ||
|
|
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
Comment on lines
+95
to
+97
| z = PhasePoint( | ||
| θ, θ, DualValue(0., θ), DualValue(0., θ) | ||
| ) |
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
| z = PhasePoint( | |
| θ, θ, DualValue(0., θ), DualValue(0., θ) | |
| ) | |
| z = PhasePoint(θ, θ, DualValue(0.0, θ), DualValue(0.0, θ)) |
| pc1 = MassMatrixAdaptor(UnitEuclideanMetric) # default dim = 2 | ||
| pc2 = MassMatrixAdaptor(DiagEuclideanMetric) | ||
| # Constructing like this until we've settled on a different interface | ||
| pc2_nutpie = NutpieVar{Float64}((2, )) |
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
| pc2_nutpie = NutpieVar{Float64}((2, )) | |
| pc2_nutpie = NutpieVar{Float64}((2,)) |
| ) | ||
| # Constructing like this until we've settled on a different interface | ||
| adaptor2_nutpie = StanHMCAdaptor( | ||
| NutpieVar{Float64}((2, )), NesterovDualAveraging(0.8, 0.5) |
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
| NutpieVar{Float64}((2, )), NesterovDualAveraging(0.8, 0.5) | |
| NutpieVar{Float64}((2,)), NesterovDualAveraging(0.8, 0.5) |
| # For this target, Nutpie (without regularization) will arrive at the true variances after two draws. | ||
| res_nutpie = runnuts_nutpie(ℓπ, DiagEuclideanMetric(D)) | ||
| @test res.adaptor.pc.var ≈ σ² rtol = 0.2 | ||
| @test preconditioned_cond(res_nutpie.adaptor.pc, Σ) < preconditioned_cond(res.adaptor.pc, Σ) |
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
| @test preconditioned_cond(res_nutpie.adaptor.pc, Σ) < preconditioned_cond(res.adaptor.pc, Σ) | |
| @test preconditioned_cond(res_nutpie.adaptor.pc, Σ) < | |
| preconditioned_cond(res.adaptor.pc, Σ) |
| # find the best preconditioner for the target. | ||
| # As these are statistical algorithms, superiority is not always guaranteed, hence this way of testing. | ||
| res_nutpie = runnuts_nutpie(ℓπ, DiagEuclideanMetric(D)) | ||
| n_nutpie_superior += preconditioned_cond(res_nutpie.adaptor.pc, Σ) < preconditioned_cond(res.adaptor.pc, Σ) |
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
| n_nutpie_superior += preconditioned_cond(res_nutpie.adaptor.pc, Σ) < preconditioned_cond(res.adaptor.pc, Σ) | |
| n_nutpie_superior += | |
| preconditioned_cond(res_nutpie.adaptor.pc, Σ) < | |
| preconditioned_cond(res.adaptor.pc, Σ) |
| return -logZ - dot(r, h.metric._temp) / 2 | ||
| #! Eq (14) of Girolami & Calderhead (2011) | ||
| function ∂H∂r( | ||
| h::Hamiltonian{<:AbstractRiemannianMetric}, θ::AbstractVecOrMat{T}, r::AbstractVecOrMat{T} |
Contributor
There was a problem hiding this comment.
[JuliaFormatter] reported by reviewdog 🐶
Suggested change
| h::Hamiltonian{<:AbstractRiemannianMetric}, θ::AbstractVecOrMat{T}, r::AbstractVecOrMat{T} | |
| h::Hamiltonian{<:AbstractRiemannianMetric}, | |
| θ::AbstractVecOrMat{T}, | |
| r::AbstractVecOrMat{T}, |
THargreaves
reviewed
Dec 20, 2025
| # println("θ: ", θ) | ||
| # println("H: ", H) | ||
| # end | ||
| G = h.metric.map(H) |
There was a problem hiding this comment.
map isn't defined for an arbitrary Riemannian metric. You'll need to introduce a layer of abstraction.
A abstract metric could have a get_G method which internally calls map if it is a Dense metric
Contributor
|
This can be closed in favour of #485, right? |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Changes include:
We should consolidate all the RHMC changes so that there is one PR to the main branch.