Visualize agglomerative collapse results

Plots entropy against number of blocks left for a collapse run. If heuristic is set to value other than NULL a second plot of the score for each merger step according to the heuristic provided is also shown.

visualize_collapse_results(sbm, heuristic = NULL, use_entropy = FALSE)

Arguments

sbm	Object of class sbm_network.
heuristic	How the best partitioning is defined. Takes either a function that takes one/two arguments: an entropy vector and an optional number of blocks vector with each element corresponding to a given location, or a string labeling algorithm. Currently only "lowest", "dev_from_rolling_mean", "delta_ratio", "trend_deviation", and "nls_residual" are supported.
use_entropy	Should the score be calculated with the entropy of the collapse? If FALSE, the entropy delta will be used. The full entropy is needed to make comparisons across separate collapses in the case of collapse_run.

Value

GGplot object comparing the fit results and each step's deviance from the rolling mean

Details

Either collapse_run, or collapse_blocks(report_all_steps = TRUE) must be run prior to calling this function. (See collapse_blocks.)

See also

Other visualizations: visualize_mcmc_trace(), visualize_network(), visualize_propensity_dist(), visualize_propensity_network()

Examples

set.seed(42)

# Start with a random networ and run agglomerative clustering with no
# intermediate MCMC steps on sbm of simulated data
net <- sim_basic_block_network(n_blocks = 3, n_nodes_per_block = 35) %>%
  collapse_blocks(sigma = 1.5)

# =============================================================================
# Visualize using no heuristic
visualize_collapse_results(net)

# =============================================================================
# Visualize using deviance from rolling average heuristic
visualize_collapse_results(net, heuristic = 'dev_from_rolling_mean')

# =============================================================================
# Visualize using custom heuristic

# Score heuristic that fits a nonlinear model to observed values and chooses by
# largest negative residual (built into package as heuristic = "nls_residual")
nls_score <- function(e, k){
  entropy_model <- nls(e ~ a + b * log(k), start = list(a = max(e), b = -25))
  -residuals(entropy_model)
}

visualize_collapse_results(net, heuristic = nls_score)