sample with LogDensityFunction: part 2 - ess.jl + mh.jl
#2590
Conversation
|
Turing.jl documentation for PR #2590 is available at: |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## py/ldf-hmc #2590 +/- ##
==============================================
- Coverage 42.32% 36.19% -6.13%
==============================================
Files 22 22
Lines 1498 1478 -20
==============================================
- Hits 634 535 -99
- Misses 864 943 +79 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
penelopeysm
force-pushed
the
py/ldf-ess
branch
2 times, most recently
from
84e8b7e to
393234b
Compare
penelopeysm
force-pushed
the
py/ldf-ess
branch
2 times, most recently
from
fa0f068 to
22c2e9a
Compare
penelopeysm
changed the title
ess.jless.jl + mh.jl
| ctx = if ldf.context isa SamplingContext | ||
| ldf.context | ||
| else | ||
| SamplingContext(rng, spl) | ||
| SamplingContext(rng, spl, ldf.context) | ||
| end |
There was a problem hiding this comment.
otherwise the existing context won't be obeyed
| # Some of the proposals require working in unconstrained space. | ||
| transform_maybe(proposal::AMH.Proposal) = proposal | ||
| function transform_maybe(proposal::AMH.RandomWalkProposal) | ||
| return AMH.RandomWalkProposal(Bijectors.transformed(proposal.proposal)) | ||
| end | ||
|
|
||
| function MH(model::Model; proposal_type=AMH.StaticProposal) | ||
| priors = DynamicPPL.extract_priors(model) | ||
| props = Tuple([proposal_type(prop) for prop in values(priors)]) | ||
| vars = Tuple(map(Symbol, collect(keys(priors)))) | ||
| priors = map(transform_maybe, NamedTuple{vars}(props)) | ||
| return AMH.MetropolisHastings(priors) |
There was a problem hiding this comment.
this code was broken for a long time and nobody bothered to update it. the idea would be that this would return an AMH.MetropolisHastings which then needed to be wrapped in an ExternalSampler. however the external sampler would break (see tests below) I just removed it because it wasn't documented and wasn't working.
| # s6 = externalsampler(MH(gdemo_default, proposal_type=AdvancedMH.RandomWalkProposal)) | ||
| # c6 = sample(gdemo_default, s6, N) | ||
|
|
||
| # NOTE: Broken because MH doesn't really follow the `logdensity` interface, but calls | ||
| # it with `NamedTuple` instead of `AbstractVector`. | ||
| # s7 = externalsampler(MH(gdemo_default, proposal_type=AdvancedMH.StaticProposal)) | ||
| # c7 = sample(gdemo_default, s7, N) |
There was a problem hiding this comment.
these were the corresponding tests
| @testset "MH link/invlink" begin | ||
| vi_base = DynamicPPL.VarInfo(gdemo_default) |
There was a problem hiding this comment.
these tests covered by requires_unconstrained_space
| check_numerical(chain, [:s, :m], [mean(InverseGamma(2, 3)), 0]; atol=0.3) | ||
| @test mean(chain[:a]) ≈ 0.0 atol = 0.1 |
There was a problem hiding this comment.
this test was annoyingly difficult to get right. with the gdemo model i had to bump this up to 50000 samples to get it to work reasonably often. I replaced it with a simpler model for which it's obvious when it's being sampled from the prior (a = 0) rather than the posterior (a = 2.5).
See #2555
Pending #2588