Convert DoodleBUGS

DoodleBUGS lets you draw a Bayesian model as a graph. mcmc convert <graph> turns a saved DoodleBUGS graph into a JuliaBUGS model file plus a fit-able spec, so a drawn model goes straight into the MCMC.js workflow.

mcmc convert model.json

wrote /path/to/model.jl
wrote /path/to/model.toml

fit it with: mcmc fit /path/to/model.toml

What it writes

From graph.json (without -o, the prefix is the graph file name without .json):

• <prefix>.jl — a JuliaBUGS model file. The graph is topologically sorted (Kahn’s algorithm) and emitted as classic BUGS model { ... } code: plates become for loops, stochastic and observed nodes become ~, deterministic nodes become <-. The generated code is wrapped in the build_model(data) contract MCMC.js expects, compiling the model with JuliaBUGS.compile.
• <prefix>.toml — a minimal, fit-able spec with backend.id = "juliabugs", pointing at the generated model file, with the data carried over from the graph.

A BUGS model must be a directed acyclic graph; if the graph contains a cycle, convert refuses rather than mis-generate.

Options

Flag	Meaning
-o, --out <prefix>	output path prefix (default: the graph file without .json)
--seed <n>	seed to write into the spec (default: 1)
--json	print the result as JSON

Then fit it

The generated spec is a normal spec file, so the rest of the workflow is unchanged:

mcmc fit model.toml -o samples.json
mcmc diagnose samples.json

The graph-to-model codegen lives in the @mcmcjs/doodleppl package, the single source of truth shared by the DoodlePPL editor and the CLI. You can also hand the graph directly to mcmc run model.json, which converts and fits in one step.