AdaptiveMultiPropTuning aborts on the first tuning step for a valid vector-backed MCMCMultiProposal.
using BAT
proposal = MCMCMultiProposal([MetropolisHastings()], [1])
algorithm = TransformedMCMC(
proposal = proposal,
proposal_tuning = AdaptiveMultiPropTuning(),
convergence = AssumeConvergence(),
nchains = 1,
)
bat_sample(
BAT.Distributions.Normal(),
algorithm,
BATContext(rng = BAT.Random123.Philox4x((0, 1))),
)On d33ddf2b:
ERROR: TaskFailedException
nested task error: UndefVarError: `curr_idx` not defined in `BAT`
@ BAT .../mcmc_adaptive_multiprop_tuner.jl:145
The method binds active_idx but passes curr_idx. Later helpers accept tuples where the public state accepts vectors, mutate tuple results, require tuple qualities although broadcasting produces vectors, reference undefined p_tuned, and alias Categorical.p.
Please standardize the tuner on vector and Categorical inputs, use active_idx, copy probability storage, and preserve normalization. Add end-to-end vector and Categorical tests for post-step and post-cycle tuning, zero-quality cases, and proposal disabling and re-enabling.
AdaptiveMultiPropTuningaborts on the first tuning step for a valid vector-backedMCMCMultiProposal.On
d33ddf2b:The method binds
active_idxbut passescurr_idx. Later helpers accept tuples where the public state accepts vectors, mutate tuple results, require tuple qualities although broadcasting produces vectors, reference undefinedp_tuned, and aliasCategorical.p.Please standardize the tuner on vector and
Categoricalinputs, useactive_idx, copy probability storage, and preserve normalization. Add end-to-end vector andCategoricaltests for post-step and post-cycle tuning, zero-quality cases, and proposal disabling and re-enabling.