Minor Touches for `ScoreGradELBO` #99

Red-Portal · 2024-09-30T17:12:00Z

No description provided.

github-actions

Benchmark Results

Benchmark suite	Current: `801a76b`	Previous: `b6083ed`	Ratio
`normal/RepGradELBO + STL/meanfield/Zygote`	`15540028767` ns	`15262008164` ns	`1.02`
`normal/RepGradELBO + STL/meanfield/ForwardDiff`	`3336745945` ns	`3126611341` ns	`1.07`
`normal/RepGradELBO + STL/meanfield/ReverseDiff`	`3252943822` ns	`3207060677` ns	`1.01`
`normal/RepGradELBO + STL/fullrank/Zygote`	`15109365150` ns	`15220603739` ns	`0.99`
`normal/RepGradELBO + STL/fullrank/ForwardDiff`	`3677404700` ns	`3439213244` ns	`1.07`
`normal/RepGradELBO + STL/fullrank/ReverseDiff`	`5745669202` ns	`5765830869` ns	`1.00`
`normal/RepGradELBO/meanfield/Zygote`	`7223742806` ns	`7257011712` ns	`1.00`
`normal/RepGradELBO/meanfield/ForwardDiff`	`2387860875` ns	`2294402627.5` ns	`1.04`
`normal/RepGradELBO/meanfield/ReverseDiff`	`1474387789` ns	`1435638209` ns	`1.03`
`normal/RepGradELBO/fullrank/Zygote`	`7231934077` ns	`7176706497` ns	`1.01`
`normal/RepGradELBO/fullrank/ForwardDiff`	`2566869136` ns	`2493123842.5` ns	`1.03`
`normal/RepGradELBO/fullrank/ReverseDiff`	`2581003358` ns	`2529829241` ns	`1.02`
`normal + bijector/RepGradELBO + STL/meanfield/Zygote`	`23549114509` ns	`23064269596` ns	`1.02`
`normal + bijector/RepGradELBO + STL/meanfield/ForwardDiff`	`10271662690` ns	`9877706080` ns	`1.04`
`normal + bijector/RepGradELBO + STL/meanfield/ReverseDiff`	`5166012174` ns	`4966628075` ns	`1.04`
`normal + bijector/RepGradELBO + STL/fullrank/Zygote`	`23512263396` ns	`23022827020` ns	`1.02`
`normal + bijector/RepGradELBO + STL/fullrank/ForwardDiff`	`10840780388` ns	`10583173397` ns	`1.02`
`normal + bijector/RepGradELBO + STL/fullrank/ReverseDiff`	`8241021709` ns	`8173149369` ns	`1.01`
`normal + bijector/RepGradELBO/meanfield/Zygote`	`14938818567` ns	`14449178740` ns	`1.03`
`normal + bijector/RepGradELBO/meanfield/ForwardDiff`	`9351470129` ns	`8845141905` ns	`1.06`
`normal + bijector/RepGradELBO/meanfield/ReverseDiff`	`3197315608` ns	`3004891758` ns	`1.06`
`normal + bijector/RepGradELBO/fullrank/Zygote`	`14961186110` ns	`14597576336` ns	`1.02`
`normal + bijector/RepGradELBO/fullrank/ForwardDiff`	`9897958470` ns	`9482666704` ns	`1.04`
`normal + bijector/RepGradELBO/fullrank/ReverseDiff`	`4617350745` ns	`4449258051` ns	`1.04`

This comment was automatically generated by workflow using github-action-benchmark.

…y_scoregradelbo

codecov · 2024-10-01T21:54:17Z

Codecov Report

Attention: Patch coverage is 95.00000% with 1 line in your changes missing coverage. Please review.

Project coverage is 92.65%. Comparing base (616b581) to head (5ff79e3).
Report is 5 commits behind head on master.

Files with missing lines	Patch %	Lines
src/objectives/elbo/scoregradelbo.jl	94.11%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #99      +/-   ##
==========================================
- Coverage   93.54%   92.65%   -0.90%     
==========================================
  Files          12       12              
  Lines         372      354      -18     
==========================================
- Hits          348      328      -20     
- Misses         24       26       +2

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Red-Portal · 2024-10-02T05:15:34Z

@willtebbutt Hi, it seems only Mooncake is failing. Could you look into the error messages? If they don't quite make sense, I'll try to pack up a MWE.

willtebbutt · 2024-10-02T08:18:15Z

Aha! I've been waiting for an example where this happens -- I've been aware of this PartialTypeVar thing for a while, but haven't had a chance to dig into it properly. I'll figure out what's going on :)

willtebbutt · 2024-10-02T08:22:43Z

This being said, if you're able to make a MWE that I can run easily on my machine, that would be great. Just whatever the function is that you're differentiating, because I'm not exactly how your tests are structured, and therefore where to find the correct function.

Sampling is now done out of the AD path.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

Red-Portal · 2024-10-04T03:42:49Z

@arnauqb Hi, I shifted things around a bit. Could you take a look and see if you have any comments? Also, the default value for the baseline, which you set as 1, is somewhat curious. Shouldn't this be 0 for the control variate to have no effect? Or maybe I'm misunderstanding something.

arnauqb · 2024-10-04T08:34:07Z

Hi @Red-Portal thanks for tidying this up, looks good to me. Yes, you are right that baseline should be 0 to have no effect, not sure what I set it to 1 😅

arnauqb · 2024-10-09T16:19:49Z

@Red-Portal After a bit of testing, I run into an error when sampling from two priors like this:

@model make_model(data)
    p1 ~ Dirichlet(3, 1.0)
    p2 ~ Dirichlet(2, 1.0)
    data ~ CustomDistribution(vcat(p1, p2))
end

and then transforming q using Bijectors.

When using the score estimator I get an input missmatch (3 !=5) error when it tries to differentiate through the scoregradelbo objective. I believe that there is something happening inside the Bijectors extension where the priors get flatten somehow for q but when passed q_stop something is different.

To fix this, we can sample samples_stop outside the gradient calculation. That is, do samples_stop = rand(rng, q_stop, obj.n_samples) and then pass samples_stop to aux, which probably makes more sense anyways.

Red-Portal · 2024-10-09T16:39:15Z

@arnauqb I think that's how it is done in this PR now. However, that error is quite peculiar, and it appears that there might be something wrong with Bijectors. If you have time, could you try to come up with a MWE and take it to Bijectors?

arnauqb · 2024-10-09T19:33:35Z

The problem may be that I am using this: https://github.com/TuringLang/Turing.jl/blob/40a0d84b76e8e262e32618f83e6b895b34177d95/src/variational/advi.jl#L23
to do the automatic transformation:

using AdvancedVI
using ADTypes
using DynamicPPL
using DistributionsAD
using Distributions
using Bijectors
using Optimisers
using LinearAlgebra
using Zygote

function wrap_in_vec_reshape(f, in_size)
    vec_in_length = prod(in_size)
    reshape_inner = Bijectors.Reshape((vec_in_length,), in_size)
    out_size = Bijectors.output_size(f, in_size)
    vec_out_length = prod(out_size)
    reshape_outer = Bijectors.Reshape(out_size, (vec_out_length,))
    return reshape_outer ∘ f ∘ reshape_inner
end

function Bijectors.bijector(
        model::DynamicPPL.Model, ::Val{sym2ranges} = Val(false);
        varinfo = DynamicPPL.VarInfo(model)
) where {sym2ranges}
    num_params = sum([size(varinfo.metadata[sym].vals, 1) for sym in keys(varinfo.metadata)])

    dists = vcat([varinfo.metadata[sym].dists for sym in keys(varinfo.metadata)]...)

    num_ranges = sum([length(varinfo.metadata[sym].ranges)
                      for sym in keys(varinfo.metadata)])
    ranges = Vector{UnitRange{Int}}(undef, num_ranges)
    idx = 0
    range_idx = 1

    # ranges might be discontinuous => values are vectors of ranges rather than just ranges
    sym_lookup = Dict{Symbol, Vector{UnitRange{Int}}}()
    for sym in keys(varinfo.metadata)
        sym_lookup[sym] = Vector{UnitRange{Int}}()
        for r in varinfo.metadata[sym].ranges
            ranges[range_idx] = idx .+ r
            push!(sym_lookup[sym], ranges[range_idx])
            range_idx += 1
        end

        idx += varinfo.metadata[sym].ranges[end][end]
    end

    bs = map(tuple(dists...)) do d
        b = Bijectors.bijector(d)
        if d isa Distributions.UnivariateDistribution
            b
        else
            wrap_in_vec_reshape(b, size(d))
        end
    end

    if sym2ranges
        return (
            Bijectors.Stacked(bs, ranges),
            (; collect(zip(keys(sym_lookup), values(sym_lookup)))...)
        )
    else
        return Bijectors.Stacked(bs, ranges)
    end
end
##

function double_normal()
    return MvNormal([2.0, 3.0, 4.0], Diagonal(ones(3)))
end

@model function normal_model(data)
    p1 ~ filldist(Normal(0.0, 1.0), 2)
    p2 ~ Normal(0.0, 1.0)
    ps = vcat(p1, p2)
    for i in 1:size(data, 2)
        data[:, i] ~ MvNormal(ps, Diagonal(ones(3)))
    end
end

data = rand(double_normal(), 100)
model = normal_model(data)

##

d = 3
μ = zeros(d)
L = Diagonal(ones(d));
q = AdvancedVI.MeanFieldGaussian(μ, L)
optimizer = Optimisers.Adam(1e-3)

bijector_transf = inverse(bijector(model))
q_transformed = transformed(q, bijector_transf)
ℓπ = DynamicPPL.LogDensityFunction(model)
elbo = AdvancedVI.ScoreGradELBO(10, entropy = StickingTheLandingEntropy()) # this doesn't
#elbo = AdvancedVI.RepGradELBO(10, entropy = StickingTheLandingEntropy()) # This works

q, _, stats, _ = AdvancedVI.optimize(
	ℓπ,
	elbo,
	q_transformed,
	10;
	adtype = AutoZygote(),
	optimizer = optimizer,
)

and stacktrace:

ERROR: output length mismatch (3 != 2)
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:35
  [2] rrule(::typeof(error), 729::String)
    @ ChainRules ~/.julia/packages/ChainRules/vdf7M/src/rulesets/Base/nondiff.jl:155
  [3] rrule(::Zygote.ZygoteRuleConfig{Zygote.Context{false}}, ::Function, ::String)
    @ ChainRulesCore ~/.julia/packages/ChainRulesCore/6Pucz/src/rules.jl:138
  [4] chain_rrule
    @ ~/.julia/packages/Zygote/Tt5Gx/src/compiler/chainrules.jl:224 [inlined]
  [5] macro expansion
    @ ~/.julia/packages/Zygote/Tt5Gx/src/compiler/interface2.jl:0 [inlined]
  [6] _pullback(ctx::Zygote.Context{false}, f::typeof(error), args::String)
    @ Zygote ~/.julia/packages/Zygote/Tt5Gx/src/compiler/interface2.jl:87
  [7] transform
    @ ~/.julia/packages/Bijectors/I8eRc/src/bijectors/stacked.jl:162 [inlined]
  [8] _pullback(::Zygote.Context{…}, ::typeof(transform), ::Stacked{…}, ::SubArray{…})
    @ Zygote ~/.julia/packages/Zygote/Tt5Gx/src/compiler/interface2.jl:0
  [9] Transform
    @ ~/.julia/packages/Bijectors/I8eRc/src/interface.jl:82 [inlined]
 [10] #84
    @ ~/.julia/packages/Bijectors/I8eRc/src/transformed_distribution.jl:221 [inlined]
 [11] #665
    @ ~/.julia/packages/Zygote/Tt5Gx/src/lib/array.jl:188 [inlined]
 [12] iterate
    @ ./generator.jl:47 [inlined]
 [13] _collect(c::Base.OneTo{…}, itr::Base.Generator{…}, ::Base.EltypeUnknown, isz::Base.HasShape{…})
    @ Base ./array.jl:854
 [14] collect_similar
    @ ./array.jl:763 [inlined]
 [15] map
    @ ./abstractarray.jl:3285 [inlined]
 [16] ∇map(cx::Zygote.Context{…}, f::Bijectors.var"#84#85"{…}, args::Base.OneTo{…})
    @ Zygote ~/.julia/packages/Zygote/Tt5Gx/src/lib/array.jl:188
 [17] adjoint
    @ ~/.julia/packages/Zygote/Tt5Gx/src/lib/array.jl:214 [inlined]
 [18] _pullback
    @ ~/.julia/packages/ZygoteRules/M4xmc/src/adjoint.jl:67 [inlined]
 [19] rand
    @ ~/.julia/packages/Bijectors/I8eRc/src/transformed_distribution.jl:218 [inlined]
 [20] _pullback(::Zygote.Context{…}, ::typeof(rand), ::Random.TaskLocalRNG, ::MultivariateTransformed{…}, ::Int64)
    @ Zygote ~/.julia/packages/Zygote/Tt5Gx/src/compiler/interface2.jl:0
 [21] estimate_scoregradelbo_ad_forward
    @ ~/code/AdvancedVI.jl/src/objectives/elbo/scoregradelbo.jl:102 [inlined]
 [22] _pullback(::Zygote.Context{…}, ::typeof(AdvancedVI.estimate_scoregradelbo_ad_forward), ::Vector{…}, ::@NamedTuple{…})
    @ Zygote ~/.julia/packages/Zygote/Tt5Gx/src/compiler/interface2.jl:0
 [23] pullback(::Function, ::Zygote.Context{false}, ::Vector{Float64}, ::Vararg{Any})
    @ Zygote ~/.julia/packages/Zygote/Tt5Gx/src/compiler/interface.jl:90
 [24] pullback(::Function, ::Vector{…}, ::@NamedTuple{…})
    @ Zygote ~/.julia/packages/Zygote/Tt5Gx/src/compiler/interface.jl:88
 [25] withgradient(::Function, ::Vector{Float64}, ::Vararg{Any})
    @ Zygote ~/.julia/packages/Zygote/Tt5Gx/src/compiler/interface.jl:205
 [26] value_and_gradient
    @ ~/.julia/packages/DifferentiationInterface/qcr6f/ext/DifferentiationInterfaceZygoteExt/DifferentiationInterfaceZygoteExt.jl:92 [inlined]
 [27] value_and_gradient!(f::Function, grad::Vector{…}, prep::DifferentiationInterface.NoGradientPrep, backend::AutoZygote, x::Vector{…}, contexts::DifferentiationInterface.Constant{…})
    @ DifferentiationInterfaceZygoteExt ~/.julia/packages/DifferentiationInterface/qcr6f/ext/DifferentiationInterfaceZygoteExt/DifferentiationInterfaceZygoteExt.jl:105
 [28] value_and_gradient!
    @ ~/.julia/packages/DifferentiationInterface/qcr6f/src/fallbacks/no_prep.jl:67 [inlined]
 [29] value_and_gradient!(ad::AutoZygote, f::Function, x::Vector{…}, aux::@NamedTuple{…}, out::DiffResults.MutableDiffResult{…})
    @ AdvancedVI ~/code/AdvancedVI.jl/src/AdvancedVI.jl:46
 [30] estimate_gradient!(rng::Random.TaskLocalRNG, obj::ScoreGradELBO{…}, adtype::AutoZygote, out::DiffResults.MutableDiffResult{…}, prob::DynamicPPL.LogDensityFunction{…}, params::Vector{…}, restructure::Optimisers.Restructure{…}, state::Nothing)
    @ AdvancedVI ~/code/AdvancedVI.jl/src/objectives/elbo/scoregradelbo.jl:134
 [31] optimize(::Random.TaskLocalRNG, ::DynamicPPL.LogDensityFunction{…}, ::ScoreGradELBO{…}, ::MultivariateTransformed{…}, ::Int64; adtype::AutoZygote, optimizer::Adam, averager::NoAveraging, show_progress::Bool, state_init::@NamedTuple{}, callback::Nothing, prog::ProgressMeter.Progress)
    @ AdvancedVI ~/code/AdvancedVI.jl/src/optimize.jl:76
 [32] optimize
    @ ~/code/AdvancedVI.jl/src/optimize.jl:50 [inlined]
 [33] #optimize#26
    @ ~/code/AdvancedVI.jl/src/optimize.jl:127 [inlined]
 [34] top-level scope
    @ ~/code/AdvancedVI.jl/score_bug.jl:97

Red-Portal · 2024-10-14T01:41:57Z

Hi @arnauqb , sorry for the late reply. Seems like this happens only on Zygote right?

Edit: Aha! This is because Zygote attempts to differentiate through rand. This should be addressed by this PR since it doesn't call rand in the ad path anymore.

test/inference/scoregradelbo_locationscale_bijectors.jl

yebai · 2024-10-23T16:01:56Z

Everything else is passing other than Enzyme tests.

arnauqb · 2024-10-28T09:06:04Z

@Red-Portal I think this line

https://github.com/TuringLang/AdvancedVI.jl/blob/45b37c111ce05f2fb01f195b8f58237bdb3a66c5/src/objectives/elbo/scoregradelbo.jl#L88C5-L88C85

should be moved outside the gradient calculation (and passed through aux). Otherwise, the gradient estimator is differentiating through the "simulator". If the simulator is not differentiable then this is just the standard score, but if it is, I think it would also be using pathwise gradients.

Red-Portal · 2024-10-28T17:29:19Z

@arnauqb

Hi! Hmm it wouldn't be using the path gradient here since samples is calculated outside. My only concern here would be the case where Zygote tries to differentiate through logpi anyways and throws an error if it can't just like when it did with rand(q). Did you experience this happen?

arnauqb · 2024-10-28T17:38:26Z

Something similar. I have a custom Distribution for which I implement a logpdf that is calculated by sampling from the distribution and measuring a loss with respect to the value passed (think it as doing generalized variational inference, if you are familiar with that). I have a custom Zygote rule for differentiating the rand of the distribution and so when calling logdensity it goes through there.

Now you are right that samples don't carry a gradient so ultimately this won't affect the results, I think, but may be a performance boost to take it out.

Red-Portal · 2024-10-28T20:07:26Z

I see. Uhghh it is annoying that Zygote can't be smarter. Anyways thanks for mentioning this, I'll try to improve it.

…y_scoregradelbo

Red-Portal · 2024-12-04T02:51:35Z

@yebai This one is also ready to go

mhauru

I don't understand the theory here, so just commenting on the software engineering. A few questions about tests, and requesting some more documentation.

src/objectives/elbo/scoregradelbo.jl

test/inference/repgradelbo_distributionsad.jl

test/inference/repgradelbo_locationscale.jl

test/inference/scoregradelbo_locationscale.jl

test/inference/scoregradelbo_locationscale_bijectors.jl

test/interface/repgradelbo.jl

mhauru · 2024-12-04T11:13:46Z

test/interface/scoregradelbo.jl

+AD_scoregradelbo_interface = if TEST_GROUP == "Enzyme"
+    [AutoEnzyme()]
+else
+    [
+        AutoForwardDiff(),
+        AutoReverseDiff(),
+        AutoZygote(),
+        AutoMooncake(; config=Mooncake.Config()),
+    ]
+end


If this gets used a lot, could it be in a test utils module? Orthogonal to this PR though, can be done later.

Yes I am planning to iron this out in a separate PR

test/interface/scoregradelbo.jl

Red-Portal · 2024-12-05T06:52:06Z

@mhauru Oh I missed a minor bug. Let me fix this tomorrow and I'll ping you again when I'm done!

Red-Portal · 2024-12-05T19:57:01Z

@mhauru Yeah seems ready to go now!

mhauru · 2024-12-06T10:29:24Z

Thanks @Red-Portal!

@mhauru Oh I missed a minor bug. Let me fix this tomorrow and I'll ping you again when I'm done!

Could something be added to the test suite that would have caught this bug?

Red-Portal · 2024-12-06T10:37:22Z

Hi @mhauru ! Nah, no need. It was a stupid typo that failed all the tests.

mhauru

I'm happy on the code side. Thanks @Red-Portal!

arnauqb · 2024-12-06T16:30:06Z

Thank you @Red-Portal , this is very useful!

fix make ScoreGradELBO immutable

00d43d0

github-actions bot reviewed Sep 30, 2024

View reviewed changes

Red-Portal added 3 commits September 30, 2024 11:13

fix error in ScoreGradELBO

af7a5a6

Merge branch 'master' of github.com:TuringLang/AdvancedVI.jl into tid…

d73ed54

…y_scoregradelbo

fix type instability, use OnlineStats for baseline window

7370248

fix default options of ScoreGradientELBO, enable more tests

1ab02c5

Red-Portal and others added 2 commits October 3, 2024 20:34

refactor ScoreGradELBO

d110ec7

Sampling is now done out of the AD path.

run formatter

42c1034

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

Red-Portal mentioned this pull request Oct 4, 2024

Type error in backward pass compintell/Mooncake.jl#282

Closed

Red-Portal added this to the v0.3.0 milestone Oct 4, 2024

fix default value for baseline control variate

feb3200

yebai added 2 commits October 22, 2024 11:28

Merge branch 'master' into tidy_scoregradelbo

45b5afd

Update CI.yml

45b37c1

yebai reviewed Oct 23, 2024

View reviewed changes

test/inference/scoregradelbo_locationscale_bijectors.jl Outdated Show resolved Hide resolved

fix move log density computation out of the AD path

86eccf7

Red-Portal added 13 commits November 4, 2024 22:42

update change the ScoreGradELBO objective to be VarGrad underneath

dc23a02

fix remove unnecessary import

f030d14

fix ScoreGradELBO outdated docs and removed unused parametric type

2577dce

update docs for ScoreGradELBO

f0bbc1b

update docs for ScoreGradELBO

43e8581

Merge branch 'master' of github.com:TuringLang/AdvancedVI.jl into tid…

3e6ef6f

…y_scoregradelbo

run formatter

9ab4d89

remove outdated test

b3d72e8

fix error in ScoreGradELBO for n_montecarlo=1

c69a5ed

add basic tests for interface tests of variational objectives

bfa9de0

run formatter

eeb2d34

tweak stepsize for inference test of ScoreGradELBO

b2fd6e2

tweak n_montecarlo in inference test

5ff79e3

Red-Portal requested a review from yebai December 4, 2024 02:51

yebai requested a review from mhauru December 4, 2024 10:38

mhauru requested changes Dec 4, 2024

View reviewed changes

Red-Portal added 2 commits December 4, 2024 15:52

add docstrings to elbo objective forward ad paths

eda4ea0

remove n_montecarlo option in the inference tests and just fix it

b6083ed

Red-Portal requested a review from mhauru December 4, 2024 20:55

Red-Portal added 2 commits December 5, 2024 14:14

fix bug in ScoreGradELBO

09c7276

fix wrong usage of n_montecarlo in inference tests

801a76b

mhauru approved these changes Dec 6, 2024

View reviewed changes

Red-Portal merged commit 1dbf2ac into master Dec 6, 2024
12 of 17 checks passed

Red-Portal deleted the tidy_scoregradelbo branch December 6, 2024 14:59

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Minor Touches for `ScoreGradELBO` #99

Minor Touches for `ScoreGradELBO` #99

Red-Portal commented Sep 30, 2024

github-actions bot left a comment •

edited

Loading

codecov bot commented Oct 1, 2024 •

edited

Loading

Red-Portal commented Oct 2, 2024

willtebbutt commented Oct 2, 2024

willtebbutt commented Oct 2, 2024

Red-Portal commented Oct 4, 2024

arnauqb commented Oct 4, 2024

arnauqb commented Oct 9, 2024

Red-Portal commented Oct 9, 2024

arnauqb commented Oct 9, 2024

Red-Portal commented Oct 14, 2024 •

edited

Loading

yebai commented Oct 23, 2024

arnauqb commented Oct 28, 2024

Red-Portal commented Oct 28, 2024 •

edited

Loading

arnauqb commented Oct 28, 2024

Red-Portal commented Oct 28, 2024

Red-Portal commented Dec 4, 2024

mhauru left a comment

mhauru Dec 4, 2024

Red-Portal Dec 4, 2024

Red-Portal commented Dec 5, 2024

Red-Portal commented Dec 5, 2024

mhauru commented Dec 6, 2024

Red-Portal commented Dec 6, 2024 •

edited

Loading

mhauru left a comment

arnauqb commented Dec 6, 2024

Minor Touches for ScoreGradELBO #99

Minor Touches for ScoreGradELBO #99

Conversation

Red-Portal commented Sep 30, 2024

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

Benchmark Results

codecov bot commented Oct 1, 2024 • edited Loading

Codecov Report

Red-Portal commented Oct 2, 2024

willtebbutt commented Oct 2, 2024

willtebbutt commented Oct 2, 2024

Red-Portal commented Oct 4, 2024

arnauqb commented Oct 4, 2024

arnauqb commented Oct 9, 2024

Red-Portal commented Oct 9, 2024

arnauqb commented Oct 9, 2024

Red-Portal commented Oct 14, 2024 • edited Loading

yebai commented Oct 23, 2024

arnauqb commented Oct 28, 2024

Red-Portal commented Oct 28, 2024 • edited Loading

arnauqb commented Oct 28, 2024

Red-Portal commented Oct 28, 2024

Red-Portal commented Dec 4, 2024

mhauru left a comment

Choose a reason for hiding this comment

mhauru Dec 4, 2024

Choose a reason for hiding this comment

Red-Portal Dec 4, 2024

Choose a reason for hiding this comment

Red-Portal commented Dec 5, 2024

Red-Portal commented Dec 5, 2024

mhauru commented Dec 6, 2024

Red-Portal commented Dec 6, 2024 • edited Loading

mhauru left a comment

Choose a reason for hiding this comment

arnauqb commented Dec 6, 2024

Minor Touches for `ScoreGradELBO` #99

Minor Touches for `ScoreGradELBO` #99

github-actions bot left a comment •

edited

Loading

codecov bot commented Oct 1, 2024 •

edited

Loading

Red-Portal commented Oct 14, 2024 •

edited

Loading

Red-Portal commented Oct 28, 2024 •

edited

Loading

Red-Portal commented Dec 6, 2024 •

edited

Loading