Module runner 

The runner: the Trainer of a Monte Carlo campaign.

Runner::run takes a Scenario and a Grid and, for every condition in order:

1. asks the callbacks whether to run it (Callback::on_condition_start);
2. builds the state with Scenario::setup;
3. runs batches of batch trials in parallel, trial t drawing from stream(seed, name, condition index, t), merges their records in trial order, and after every batch evaluates the stopping rule and calls Callback::on_batch_end; after the first batch it checks that the targets of the rule are recorded (Stop::check_metrics), and the last batch ends at the bound of the rule on the trials (Stop::trial_limit);
4. hands the ConditionResult to the callbacks and the loggers.

The results are identical whatever the number of threads (see crate). The batch size only sets how often the rule is checked (a larger batch keeps every core busy, a smaller one stops closer to an error or precision target); a bound on the trials is met exactly, so that max_trials(1000) runs 1000 trials with batches of 32 as with batches of 1. Every condition runs at least one trial.

Examples

use comnum_sim::prelude::*;
use comnum_mapping::{constellation::{Constellation, ConstellationKind}, source as sources};
use comnum_channel::awgn as channels;
use comnum_metrics as metrics;
use comnum_sim::scenario::from_fn;
let coin = from_fn("coin", |cond, rng, rec| {
    let p = cond.f64("p")?;
    rec.errors("heads", (rng.uniform() < p) as u64, 1);
    Ok(())
});
let grid = Grid::new().axis("p", [0.1, 0.5]);
let one = Runner::new().seed(3).threads(1).batch(50).stop(Stop::max_trials(1000))
    .run(&coin, &grid).unwrap();
let four = Runner::new().seed(3).threads(4).batch(50).stop(Stop::max_trials(1000))
    .run(&coin, &grid).unwrap();
assert_eq!(one[0].metric("heads"), four[0].metric("heads"));   // thread-count independent
assert!((one[1].metric("heads").unwrap().value() - 0.5).abs() < 0.05);

Structs

ConditionResult

The result of one condition.

RunInfo

What the callbacks and loggers know about the run.

RunReport

The summary of a finished run.

Runner

The campaign runner (builder; see the module documentation).

Enums

Status

How a condition ended.

Functions

metric_columns

The names of the columns of the metric name in a flat row (ConditionResult::row): name, name_lo and name_hi (the estimate and its confidence interval), name_count and name_total (the errors and the events of a rate, the number of values of a mean twice).