Sweep a technology family's clusterings, to choose k by looking

Clusters one family over a range of k and reports what each one buys: within-cluster dispersion, average silhouette and the cluster sizes. Nothing is changed – the point is to see where the structure is before naming a k in aggregate_model_regions()(clusters = ).

Usage

process_cluster_sweep(
  object,
  group,
  ks = NULL,
  geoscale = NULL,
  level,
  features = NULL
)

Arguments

object

a model.

group

a process group name, as get_process_groups() reports it.

ks

integer vector of cluster counts. NULL (default) sweeps the whole admissible range: from one cluster per level region to one per region.

geoscale

a geoscales::GeoScale; defaults to the model's own.

level

the target geoframe, e.g. "zone".

features

feature spec, as in aggregate_model_regions().

Value

A data.frame, one row per k, with the dispersion and silhouette columns nestedscales::cluster_sweep() returns plus sizes, the cluster sizes as a comma-separated string.

Details

The sweep uses the same features and the same contiguity graph the aggregation would, so what you inspect is what you would get. Two families will generally disagree about where their structure is, which is why k is set per family and not once for the model.

Read it as a scree plot: dispersion falls with k and always will, so take the knee rather than the minimum, and prefer a k whose silhouette is high and whose sizes are not one big cluster and a row of singletons. There is deliberately no "best k" here, and none in nestedscales either – the number depends on what the model is for.

See also

get_process_groups(), aggregate_model_regions()

Other geoscale: check_geoscale_regions(), get_process_groups(), model_clusters(), plot_geoscale(), plot_map(), setCalendar,config-method

Examples

if (FALSE) { # \dontrun{
ek  <- topia$modules$electricity$R11
mod <- newModel("TOPIA", data = ek$repo,
                calendar = topia$modules$calendars$s4_h24,
                region = ek$regions, horizon = topia$modules$horizons$base)
mod <- setGeoscale(mod, ek$geoscale)

# how a family's regions group at each k, and the spread left inside the
# clusters -- the input to choosing `clusters` for
# aggregate_model_regions(). The family needs parameters that actually
# differ by region (site yields, site costs): one value for every region
# leaves nothing to cluster on.
process_cluster_sweep(mod, group = "EWIN", ks = 3:6,
                      geoscale = topia$geoscales$honeycomb, level = "zone")
} # }