Skip to content

Run files and CLI

A YAML run file describes a model and what to do with it. Everything in Models and priors and Samplers has a run-file form.

run.yaml
theory:
  W0WaCDM: {rdrag: fit, m_nu: 0.06}

likelihood:
  DESIDR2BAO: null
  PlanckPR4EarlyLCDM: null

params:
  h: {prior: {min: 0.5, max: 0.9}, ref: 0.66, latex: h}
  omega_b: {prior: {min: 0.018, max: 0.026}, ref: 0.0222}
  omega_cdm: {prior: {min: 0.08, max: 0.16}, ref: 0.119}
  w0: {prior: {min: -3.0, max: 1.0}, ref: -0.8}
  wa: {prior: {min: -3.0, max: 2.0}, ref: -0.7}
  Omega_m:                                  # derived: a Defined parameter
    value: "lambda h, omega_b, omega_cdm: (omega_b + omega_cdm + 0.06 / 93.14) / h**2"
    latex: '\Omega_\mathrm{m}'

sampler:
  nuts: {num_warmup: 1000, num_samples: 2000, num_chains: 4}

maximize: {num_starts: 4}
profile: {params: {w0: {min: -1.4, max: -0.2, n: 41}}}

output: chains/run
seed: 0
  • A bare number fixes a parameter. prior takes {min, max} or {dist: norm, loc, scale}.
  • ref is the start point. latex is the label.
  • A value lambda defines a parameter from others, as Defined does.
  • Unknown theory and likelihood names are import paths (module:attribute), so your own classes work.
  • Several files merge left to right, so a base file plus a variant is two arguments.

Commands

Command Does
cosmonuts install NAME... [--all] [--path P] Download datasets
cosmonuts sample run.yaml Sample with the sampler block
cosmonuts maximize run.yaml Best fit, written to <output>.bestfit.yaml
cosmonuts profile run.yaml Profile likelihood
cosmonuts plot triangle\|trace\|profile\|stats ... Figures and tables from saved chains
cosmonuts info theories\|likelihoods\|samplers What a run file can name and set 
cosmonuts sample run.yaml --test                      # build, evaluate once, stop
cosmonuts sample run.yaml                             # run
cosmonuts sample run.yaml sampler.nuts.num_samples=200 seed=3   # key.path=value overrides
cosmonuts sample run.yaml --resume                    # continue; --force overwrites

From Python, config.run does the same as cosmonuts sample. --device cpu|gpu, --seed and --output override the file.