samOptiPro — Systematic Optimization of Bayesian Stock Assessment Models in NIMBLE
A workflow-driven helper to configure, assess, and optimise MCMC sampling in NIMBLE, following a reproducible decision tree:
general tools → assess → detect poor performance → identify bottlenecks →
(model surgery | custom samplers) → reassess → validate → iterate.
samOptiPro provide an advanced, modular workflow for diagnosing, benchmarking, and optimizing hierarchical ecological models (SAM-like frameworks) built in NIMBLE.
It integrates:
- Structural diagnostics: detect non-differentiable nodes (truncations, Dirichlet, simplex constraints…)
- Adaptive sampler configuration: auto-assign Slice / AF_slice / RW / Block / HMC / NUTS
- Performance analytics: algorithmic (ESS, ESS/s) and computational (runtime) efficiency
- Automatic visual reports: bottlenecks, convergence, and Rhat distributions
- Differentiability testing: seamless handoff to gradient-based inference (nimbleHMC)
- Hybrid family strategy: use block samplers or adaptive HMC where correlations demand it
# From local development directory
devtools::load_all("samOptiPro")
# Or (once hosted)
# remotes::install_github("RomualdEcoStats/samOptiPro-core")