Extracts the summary information recorded for the individual algorithm runs
performed by blvim().
Arguments
- sim
A spatial interaction model object produced by
blvim(), i.e. an object of classsimandsim_blvim
Value
A data frame of class blvim_logs with one row per starting point
and the following columns:
run: index of the run, from 1 to the number of starting points;potential: final value of the potential;iterations: number of iterations;converged: whether the run converged;status: stopping status reported by a gradient algorithm;gap: final duality gap reported by a gradient algorithm;evaluations: number of evaluations reported by a gradient algorithm;n_active: number of active destinations of the final configuration;configuration: identifier of the final configuration (see details);selected: whether the run is the one returned byblvim().
The gradient-specific columns are NA for the "linear" and
"quadratic" algorithms.
Details
The rows are in the same order as the starting points supplied to
blvim(). Run summaries are retained independently of the value of
keep_runs.
The final configuration of a run is described by its rescaled attractivenesses \(U_j = \kappa_j Z_j / \sum_k \kappa_k Z_k\). A destination is active when \(U_j > 10^{-5}\). Two runs reach the same configuration when they have the same active destinations and their rescaled attractivenesses differ by at most \(10^{-3}\). The configurations are numbered by decreasing best potential, so that configuration 1 is the one of the selected run. The number of runs that reach it measures the reliability of the multiple starts: when it is reached by a single run, more starting points might find a better one.
The summary() and autoplot() methods of the result (see
summary.blvim_logs() and autoplot.blvim_logs()) describe and display
these summaries.
See blvim() for details on how the nature of Z determines the number of
runs. Use sim_runs() to extract the complete spatial interaction models
produced by the runs when they are available.