Each scenario/iteration/MP has independent OM, observation, and EM state. Environmental random seeds are shared across MPs, not their realised data. Separate EM seeds prevent stochastic estimators affecting the OM stream. Results contain one row per annual decision, plus a manifest and final states.
Usage
run_mse(
om,
mps,
scenarios,
iterations,
years,
seed = 271828L,
cores = 1L,
checkpoint_dir = NULL,
resume = TRUE,
rng = c("keyed", "cra2_legacy"),
max_steps = Inf,
progress = NULL
)Arguments
- om
An
mse_om()orom_decamod()adapter.- mps
A list of
mp()objects, with unique identifiers.- scenarios
Unique scenario identifiers understood by the OM adapter.
- iterations
Positive integer iteration identifiers, not a count.
- years
Consecutive modelled projection years.
- seed
Non-negative integer master seed.
- cores
Worker count. Fork parallelism is available on Unix; other platforms use serial execution. Default one avoids an unexpected large run.
- checkpoint_dir
Optional directory for annual RDS checkpoints.
- resume
Whether to resume matching checkpoints.
- rng
"keyed"uses deterministic seeds keyed by scenario/iteration/year."cra2_legacy"reproduces the toy seed formula for numerical regression tests; it requires positive integer scenario IDs and rejects designs with overlapping scenario/iteration seeds.- max_steps
Maximum number of additional scenario/year steps this call. Useful for smoke tests; the unchanged design can subsequently resume.
- progress
Optional function receiving the newly completed annual rows.