opticskxi (version 1.2.0)

ensemble_models: Select best models based on ensemble metrics

Description

Call ensemble_metrics with varying numbers of rank thresholds to consider and sum up the ranks from those calls.

Usage

ensemble_models(
  df_kxi,
  n_models = 4,
  metrics = NULL,
  metrics_exclude = NULL,
  model_subsample = c(0.1, 0.2, 0.5),
  n_models_subsample = 10
)

Value

Input object df_kxi subsetted to best models according to ensemble metrics.

Arguments

df_kxi

Output of opticskxi_pipeline function. Dataframe with models' parameters and OPTICS k-Xi results

n_models

Number of best models to return

metrics

Names of metrics to use. Any of those computed by opticskxi_pipeline, e.g. 'sindex', 'ch', 'dunn', 'dunn2', 'widestgap', 'entropy' etc. NULL for all (8).

metrics_exclude

Names of metrics to exclude. Typically used with metrics = NULL. E.g. 'entropy'.

model_subsample

Ratios of best models to consider.

n_models_subsample

Number of best models when subsampling.

Examples

data('m_psych_embeds')
m_psych_embeds = m_psych_embeds[1:200, 1:20]

df_params = expand.grid(n_xi = 4:5, pts = c(5, 10), dist = 'cosine',
                        dim_red = 'ICA', n_dimred_comp = 5)

df_kxi = opticskxi_pipeline(m_psych_embeds, df_params,
                            metrics_dist = 'cosine',
                            n_min_clusters = 2, n_cores = 1,
                            metrics = c('avg.silwidth', 'dunn'))

df_kxi = ensemble_models(df_kxi, n_models = 4,
                         model_subsample = c(0.4, 0.6),
                         n_models_subsample = 4)