Skip to contents

Summary of the runs of a multiple start blvim

Source: R/blvim_logs.R
summary.blvim_logs.Rd

Summarises the runs of blvim() described by sim_logs(): convergence of the runs, cost of the runs, and distinct final configurations reached by the runs.

Usage

# S3 method for class 'blvim_logs'
summary(object, ...)

# S3 method for class 'summary_blvim_logs'
print(x, ...)

Arguments

object

a data frame of class blvim_logs produced by sim_logs()

...

additional parameters (not used currently)

x

an object of class summary_blvim_logs produced by summary.blvim_logs()

Value

an object of class summary_blvim_logs and list with the following elements:

  • n_runs: number of runs;

  • n_converged: number of runs that converged;

  • status: number of runs for each status reported by the gradient algorithms (NULL for the other algorithms);

  • iterations, evaluations: minimum, median and maximum number of iterations and of evaluations of the runs (evaluations is NULL for algorithms that do not report it);

  • max_gap: largest final duality gap of the converged runs (NA when not reported);

  • configurations: a data frame with one row per final configuration, by decreasing potential, with columns configuration, potential (best potential of its runs), n_active, runs (number of runs reaching it), share (proportion of the runs) and converged (number of converged runs reaching it);

  • selected_converged: whether the selected run converged.

Details

The final configurations are identified as explained in sim_logs(). The number of runs that reach the configuration of the selected run (the best one) measures the reliability of the multiple starts. When this configuration is reached by a single run, a better one might have been missed and more starting points are advisable.

See also

sim_logs(), blvim(), autoplot.blvim_logs()

Examples

distances <- french_cities_distances[1:15, 1:15] / 1000
production <- rep(1, 15)
set.seed(1)
starts <- blvim_initialise(n = 20)
model <- blvim(distances, production, 1.5, 1 / 50, starts,
  algorithm = "gradient", bipartite = FALSE
)
summary(sim_logs(model))
#> 
#> ── Summary of 20 runs of blvim ──
#> 
#> • 20 of 20 runs converged.
#> • Status: converged (20).
#> • Iterations: median 22, range [16, 68].
#> • Evaluations: median 34, range [25, 91].
#> • Largest duality gap of the converged runs: 1.574e-05.
#> 
#> ── Final configurations 
#> 16 distinct configurations. The best one (potential -13.42567, 13 active
#> destinations) was reached by 3 of 20 runs.
#> The second best potential is -13.42744 (difference 0.001761).
#>  configuration potential n_active runs share converged
#>              1  -13.4257       13    3  0.15         3
#>              2  -13.4274       13    1  0.05         1
#>              3  -13.4328       13    2  0.10         2
#>              4  -13.4334       13    1  0.05         1
#>              5  -13.6183       12    1  0.05         1
#> ... and 11 more.