R/bootThreshold.R

In bootnet: Bootstrap Methods for Various Network Estimation Routines

# Function to threshold network based on bootstrap samples:
bootThreshold <- function(bootobject, alpha = 0.05,verbose=TRUE, thresholdIntercepts = FALSE){
  # Check if object is bootnet object:
  if (!is(bootobject,"bootnet")){
    stop("'bootobject' must be an object of class 'bootnet'")
  }

  # Check type:
  if (bootobject$type != "nonparametric" & bootobject$type != "parametric"){
    stop("Bootstrap type must be 'nonparametric' or 'parametric'")
  }

  # Check alpha:
  if (verbose){
    exp <- expAlpha(alpha,length(bootobject$boots))
    message(paste0("Expected significance level given number of bootstrap samples is approximately: ",format(signif(exp,2),scientific = FALSE)))
  }


  # Extract the network object:
  Network <- bootobject$sample
  # Dummy for multiple graphs:
  if (!is.list(Network$graph)){
    Graphs <- list(Network$graph)
    Directed <- list(Network$directed)
    Intercepts <- list(Network$intercepts)
    names(Graphs) <- names(Directed) <- names(Intercepts) <-
      unique(bootobject$bootTable$graph)
  } else {
    Graphs <- Network$graph
    Directed <- Network$directed
    Intercepts <-  Network$intercepts
  }

  # For every graph:
  for (g in seq_along(Graphs)){
    graphName <- names(Graphs)[g]

    # Summary table of edge weights:
    bootSummary <- bootobject$bootTable %>%
      dplyr::filter(.data[['type']] == "edge", .data[['graph']] == graphName) %>%
      dplyr::group_by(.data[['node1']],.data[['node2']]) %>%
      dplyr::summarize(
        lower = quantile(value, alpha/2),
        upper = quantile(value, 1 - alpha/2)
      ) %>%
      dplyr::mutate(sig = upper < 0 | lower > 0) %>%
      filter(!.data[['sig']])

    # Threshold network:
    if (nrow(bootSummary) > 0){
      for (i in 1:nrow(bootSummary)){
        Graphs[[graphName]][Network$labels == bootSummary$node1[i],Network$labels == bootSummary$node2[i]] <- 0
        if (!Directed[[graphName]]){
          Graphs[[graphName]][Network$labels == bootSummary$node2[i],Network$labels == bootSummary$node1[i]] <- 0
        }
      }
    } else {
      message("All edges indicated to be nonzero")
    }

    # Threshold intercepts
    if (thresholdIntercepts){

      # Summary table of edge weights:
      bootSummary <- bootobject$bootTable %>%
        dplyr::filter(.data[['type']] == "intercept", .data[['graph']] == graphName) %>%
        dplyr::group_by(.data[['node1']]) %>%
        dplyr::summarize(
          lower = quantile(value, alpha/2),
          upper = quantile(value, 1 - alpha/2)
        ) %>%
        dplyr::mutate(sig = upper < 0 | lower > 0) %>%
        filter(!.data[['sig']])

      # Threshold network:
      if (nrow(bootSummary) > 0){
        for (i in 1:nrow(bootSummary)){
          Intercepts[[graphName]][Network$labels == bootSummary$node1[i]] <- 0
        }
      } else {
        message("All intercepts indicated to be nonzero")
      }
    }
  }

  # Return to network object:
  if (length(Graphs) == 1){
    Network$graph <- Graphs[[1]]
    Network$intercepts <- Intercepts[[1]]
  } else {
    Network$graph <- Graphs
    Network$intercepts <- Intercepts
  }

  # Add indicator network is thresholded:
  Network$thresholded <- TRUE

  # Return network:
  return(Network)
}