R/impute.network.list.R

In tergm: Fit, Simulate and Diagnose Models for Network Evolution Based on Exponential-Family Random Graph Models

#  File R/impute.network.list.R in package tergm, part of the
#  Statnet suite of packages for network analysis, https://statnet.org .
#
#  This software is distributed under the GPL-3 license.  It is free,
#  open source, and has the attribution requirements (GPL Section 7) at
#  https://statnet.org/attribution .
#
#  Copyright 2008-2024 Statnet Commons
################################################################################


#' Impute missing dyads in a series of networks
#' 
#' This function takes a list of networks with missing dyads and returns a list
#' of networks with missing dyads imputed according to a list of imputation
#' directives.
#' 
#' 
#' @param nwl A list of [`network`] objects or a
#'   [`network.list`] object.
#' @param imputers A character vector giving one or more methods to
#'   impute missing dyads. Currenly implemented methods are as
#'   follows: \describe{ \item{`next`}{Impute the state of the same
#'   dyad in the next network in the list (or later, if that one is
#'   also missing).  This imputation method is likely to lead to an
#'   underestimation of the tie-change rates. The last
#'   network in the list cannot be imputed this way.}
#'   \item{`previous`}{Impute the state of the same dyad in
#'   the previous network in the list (or earlier, if that one is also
#'   missing). The first network in the list cannot be imputed this
#'   way.}  \item{`majority`}{Impute the missing dyad with
#'   the value of the majority among the non-missing dyads in that
#'   time step's network. A network that has exactly the same number
#'   of ties as non-missing non-ties cannot be imputed this way.}
#'   \item{`0`}{Assume missing dyads are all non-ties.}
#'   \item{`1`}{Assume missing dyads are all ties.} } If
#'   \code{length(imputers)>1} the specified imputation methods will
#'   be applied in succession. For example,
#'   \code{imputers=c("next","previous","majority","0")} would first
#'   try to impute a missing dyad with the next time step's value. If
#'   it, and all of the later values for that dyad are missing, it
#'   will try to impute it with the previous time step's value. If it,
#'   and all of the earlier values for that dyad are missing as well,
#'   it will try to impute it with the value of the majority of
#'   non-missing dyads for that time step. If there is an exact tie,
#'   it will impute 0.
#' @param nwl.prepend An optional list of networks to treat as
#'   preceding those in \code{nwl}. They will not be imputed or
#'   returned, but they can be useful for imputing dyads in the first
#'   network in \code{nwl}, when using \code{"previous"} imputer.
#' @param nwl.append An optional list of networks to treat as
#'   following those in \code{nwl}. They will not be imputed or
#'   returned, but they can be useful for imputing dyads in the last
#'   network in \code{nwl}, when using \code{"next"} imputer.
#' @return A list of networks with missing dyads imputed.
#' @seealso [`network`], [is.na()]
#' @keywords manip
#' @export impute.network.list
impute.network.list <- function(nwl, imputers=c(), nwl.prepend=list(), nwl.append=list()){
  # TODO: Make it possible to write one's own imputers. E.g., break the following out into impute.network.stop(), impute.network.next(), etc..
  IMPUTERS <- c("next", "previous", "majority", "0", "1")
  imputers <- IMPUTERS[pmatch(imputers,IMPUTERS)]
  if(any(is.na(imputers))) stop("Unknown imputation option: ",sQuote(imputer),". Impute options must be a character vector containing one or more of ", dQuote("next"), ", ", dQuote("previous"), ", ", dQuote("majority"), ", ", dQuote("0"), ", and/or ", dQuote("1"), ".")
  
  for(imputer in imputers){
    nwl <- switch(imputer,
                  previous = {
                    nwl <- c(nwl.prepend, nwl)
                    nwl.NA <- sapply(nwl, network.naedgecount)>0 # Update which networks have missing dyads.
                    
                    for(t in seq_along(nwl)[-1])
                      if(nwl.NA[t]){
                        # Workaround for a bug in network (Ticket #80 in Trac)
                        na.el <-as.edgelist(is.na(nwl[[t]]))
                        na.eids <- apply(na.el, 1, function(e) get.edgeIDs(nwl[[t]], e[1],e[2], na.omit=FALSE))
                        nwl[[t]] <- delete.edges(nwl[[t]], na.eids)
                        nwl[[t]][na.el] <- nwl[[t-1]][na.el]
                      }
                    # Remove the prepended networks.
                    nwl[length(nwl.prepend)+seq_len(length(nwl)-length(nwl.prepend))]
                  },
                  `next` = {
                    nwl <- c(nwl, nwl.append)
                    nwl.NA <- sapply(nwl, network.naedgecount)>0
                    
                    for(t in rev(seq_along(nwl)[-length(nwl)]))
                      if(nwl.NA[t]){
                        # Workaround for a bug in network (Ticket #80 in Trac)
                        na.el <-as.edgelist(is.na(nwl[[t]]))
                        na.eids <- apply(na.el, 1, function(e) get.edgeIDs(nwl[[t]], e[1],e[2], na.omit=FALSE))
                        nwl[[t]] <- delete.edges(nwl[[t]], na.eids)
                        nwl[[t]][na.el] <- nwl[[t+1]][na.el]
                      }
                    # Remove the appended networks.
                    nwl[seq_len(length(nwl)-length(nwl.append))]
                  },
                  majority = {
                    lapply(nwl, function(y){
                      impute <- sign(network.edgecount(y,na.omit=TRUE)/network.dyadcount(y,na.omit=TRUE)-0.5)
                      if(impute==0){# If exact tie, can't impute.
                        y
                      }else{
                        # Workaround for a bug in network (Ticket #80 in Trac)
                      na.el <-as.edgelist(is.na(y))
                      na.eids <- apply(na.el, 1, function(e) get.edgeIDs(y, e[1],e[2], na.omit=FALSE))
                        impute <- impute > 0
                        y <- delete.edges(y, na.eids)
                        y[na.el] <- impute
                        y
                      }
                    })
                  },
                  `0` = {
                    lapply(nwl, function(y){
                      # Workaround for a bug in network (Ticket #80 in Trac)
                      na.el <-as.edgelist(is.na(y))
                        na.eids <- apply(na.el, 1, function(e) get.edgeIDs(y, e[1],e[2], na.omit=FALSE))
                      y <- delete.edges(y, na.eids)
                    })
                  },
                  `1` = {
                    lapply(nwl, function(y){
                      # Workaround for a bug in network (Ticket #80 in Trac)
                      na.el <-as.edgelist(is.na(y))
                      na.eids <- apply(na.el, 1, function(e) get.edgeIDs(y, e[1],e[2], na.omit=FALSE))
                      y <- delete.edges(y, na.eids)
                      y[na.el] <- 1
                      y
                    })
                  }
                  )
  }

  nwl.NA <- sapply(nwl, network.naedgecount)>0
     
  if("previous" %in% imputers && nwl.NA[1]) warning("Imputation option `previous' cannot impute dyads of the first network in the series.")
  if("next" %in% imputers && nwl.NA[length(nwl.NA)]) warning("Imputation option `next' cannot impute dyads of the last network in the series.")
  if("majority" %in% imputers && any(nwl.NA)) warning("Imputation option `majority' encountered an exact tie.")
  
  nwl  
}