Learn Configuration

specs: ../03-sample/specs.yaml
models:
  rdf:
    model_path: ../05-lgp/models/rdf.lgp
    tolerance: 0.1
  pmf:
    model_path: ../05-lgp/models/pmf.lgp
    n_eff: 5
mcmc:
  total_steps: 10000
  warmup: 2000
  thin: 1
  resume: false
  device: cuda
output:
  directory: ./
  overwrite: false

Each model selects exactly one effective-observation mode: positive n_eff, independent_observations: true, or a positive curve tolerance. MCMC options also include priors_disttype, progress_stride, n_walkers, rhat_tol, ess_min, and include_implicit_charge.

Learning exposes one output root and owns these fixed paths:

learn.log
plots/
  marginals.pdf
  qoi-marginals.pdf
  corner.pdf
outputs/
  specs.yaml
  prior.pt
  posterior.pt
  mcmc.ckpt

Existing owned files are rejected by default. output.overwrite: true removes only the paths listed above. mcmc.resume: true requires a checkpoint, regenerates posterior and plots, and appends a delimited run to learn.log. Resume and overwrite cannot be combined.

The configured specs.yaml is copied unchanged into outputs/. The prior, posterior, and checkpoint are written atomically. Resume requires the copied specifications and validates the specification fingerprint, ordered models and their hashes, target settings, dimensions, walkers, warmup, thinning, prior family, and proposal. All three artifacts and all three plots are mandatory for command success. Marginal annotations show posterior means directly below their lower bounds.