Custom GJAM Functions

These are some functions I use to prettify GJAMTime output. I’ll add inline examples later but for now here are some screenshots.

Basic plot theme

basicPlotTheme <- function() {
  # Some basic plot attributes to make ggplot more readable
  th <- theme(axis.text.x = element_text(angle=45, vjust=0.5) ,
        text = element_text(size=18),
        title = element_text(size=24) )
  return(th)
}

Alphas

extractAlphasTall <- function(gjam.output) {
  # returns a tall version of interaction values, each row being one
  # species-by-species interaction value. dim S^2 x 3 with columns 
  # "to" and "from" species, and the estimated interaction coefficient.
  a <- gjam.output$parameters$alphaTable
  spp <- colnames(out$modelList$effort$values)
  ns <- length(spp)
  pairs.df <- data.frame(to=rep(spp, each=ns),
                         from=rep(spp, by=ns),
                         est=0, logest=NA, sign=NA)
  pairs <- unlist(str_split(a[,1], ', '))
  to <- pairs[seq(1, length(pairs), 2)]
  from <- pairs[seq(2, length(pairs), 2)]
  
  for (i in 1:nrow(a)) {
    est <- a$Estimate[i]
    inx <- (pairs.df$to==to[i]) + (pairs.df$from==from[i]) == 2
    pairs.df[inx, 3:ncol(pairs.df)]<- c(est, log(abs(est)), est>=0)
  }
  return(pairs.df)
}

extractAlphasSquare <- function(gjam.output) {
  # returns the S x S matrix of species interaction values
  a <- gjam.output$parameters$alphaTable
  spp <- colnames(out$modelList$effort$values)
  ns <- length(spp)
  square <- data.frame(matrix(0, nrow=ns, ncol=ns))
  colnames(square) <- rownames(square) <- spp
  pairs <- unlist(str_split(a[,1], ', '))
  to <- pairs[seq(1, length(pairs), 2)]
  from <- pairs[seq(2, length(pairs), 2)]
  for (i in 1:nrow(a)) { square[ from[i], to[i] ] <- a$Estimate[i] }
  return(square)
}

alphaPlot <- function(gjam.output, is.log=F) {
  # Plots species-by-species interaction matrix, with option to 
  # take natural log of values.
  pairs.df <- extractAlphasTall(gjam.output)
  title <- 'species interactions'
  if (is.log) {
    title <- paste("log of", title)
    alpha.plot <- pairs.df %>% ggplot() + geom_tile(aes(x=to, y=from, fill=logest))
  } else {
    alpha.plot <- pairs.df %>% ggplot() + geom_tile(aes(x=to, y=from, fill=est))
  }
  alpha.plot <- alpha.plot +
    ggtitle(title) +
    scale_fill_distiller(type='div', palette=4) +
    basicPlotTheme() + theme(axis.text.x = element_text(hjust=1, vjust=1))
  alpha.plot
}

Correlation plots

corPlot <- function(ydata) {
  # Plots species correlations directly from the ydata.
  cor.plot <- round(cor(ydata), 4) %>%
    melt() %>%
    rename(Species1="Var1", Species2="Var2", correlation='value') %>%
    ggplot(aes(x=Species1, y=Species2, fill=correlation)) + 
    geom_tile() +
    ggtitle('species correlations') +
    scale_fill_gradient2(low = "white", high = "orange", midpoint = 0.5) +
    basicPlotTheme() + theme(axis.text.x = element_text(hjust=1, vjust=1))
  cor.plot
}

Sensitivity plots

sensPlot <- function(gjam.output, is.horiz=T, is.ordered=F) {
  # Plots sensitivity output from GJAMTime; options to order by sensitivity
  # value or to flip axes horizontally (default) or vertically.
  sens.plot <- gjam.output$parameters$sensTable %>%
    mutate(covariates = gsub(':', ':\n', rownames(.)),
           interaction = ifelse(grepl(':', covariates), 'true', 'false'))
  if (is.ordered) {
    sens.plot <- sens.plot %>%
      mutate(covariates = fct_reorder(covariates, Estimate, .fun='median'))
  }
  sens.plot <- sens.plot %>%
    ggplot(aes(x=reorder(covariates, Estimate), color=interaction)) +
    geom_point(aes(y=Estimate)) +
    geom_errorbar(aes(ymax=Estimate+SE, ymin=Estimate-SE), position="dodge") +
    ggtitle('community sensitivity to betas') + xlab('covariate') + ylab('sensitivity estimate') +
    scale_color_brewer(type='qual', palette=2) + basicPlotTheme()
  if (is.horiz) adjust <- coord_flip() 
  else adjust <- theme(axis.text.x = element_text(hjust=1, vjust=1))
  sens.plot + adjust
}