In neurodata/r-mgc: Multiscale Graph Correlation

require(ggplot2)
require(reshape2)
require(grid)
require(gridExtra)
require(mgc)
require(ICC)
require(I2C2)
require(cowplot)
require(dplyr)
require(tidyr)

fMRI Figure

fmri.results <- readRDS('./data/real/rf_fmri_results.rds')
stat.results <- fmri.results$statistics
problem.results <- fmri.results$rf

#stat.results <- subset(stat.results, alg != "I2C2")
stat.results.sum  <- stat.results %>%
  group_by(thresh, alg, nroi) %>%
  summarize(stat=weighted.mean(stat, nsub), nsub=sum(nsub), nscans=sum(nscans))
problem.results.sum  <- problem.results %>%
  group_by(thresh, task, Metric, nroi, embed) %>%
  summarize(stat=weighted.mean(stat, nsub), null=weighted.mean(null, nsub), nsub=sum(nsub), nscans=sum(nscans))

max.stat <- stat.results.sum %>%
  group_by(alg) %>%
  slice(which.max(stat))

min.prob <- problem.results.sum %>%
  group_by(embed, task) %>%
  slice(which.min(stat))

problem.results.sum %>%
  gather("model", "statistic", -thresh, -task,  -Metric, -nroi, -nsub, -nscans, -embed) %>%
  ggplot() +
    geom_line(aes(color=task, linetype=hypothesis, x=thresh, y=statistic)) +
    geom_line(data=stat.results.sum, aes(color=alg, x=thresh, y=stat)) +
    geom_point(data=min.prob, aes(x=thresh, y=stat, color=task)) +
    geom_point(data=max.stat, aes(x=thresh, y=stat, color=alg)) +
    ggtitle("Examining Task Performance with RF Classification") +
    facet_wrap(.~embed) +
    theme_bw() +
    xlab("Threshold") +
    ylab("Statistic")

class.res <- subset(classifier.results, k == 4 & Metric %in% c("RMSE", "Accuracy") & task != "Lifestyle")
class.res$value[class.res$Metric == "Accuracy"] <- 1 - class.res$value[class.res$Metric=="Accuracy"]
min.task <- data.frame(min.thresh=c(), min.stat=c(), task=c())
for (un in unique(class.res$task)) {
  ss <- class.res[class.res$task == un,]
  ss$value <- (ss$value - min(ss$value))/(max(ss$value) - min(ss$value))
  class.res$value[class.res$task == un] <- ss$value
  min.pt <- which(ss$value == min(ss$value))[1]
  min.task <- rbind(min.task, data.frame(min.thresh=ss$thresh[min.pt], min.stat=ss$value[min.pt], task=un))
}

ggplot(stat.results, aes(x=thresh, color=alg, y=stat, group=alg)) +
  geom_line() +
  theme_bw() +
  geom_line(data=class.res, aes(x=thresh, y=value, color=task, group=task), size=1, linetype=6) +
  geom_point(data=max.alg, aes(x=max.thresh, y=max.stat, color=alg, group=alg), size=2, show.legend=FALSE) +
  geom_vline(data=max.alg, aes(xintercept=max.thresh, color=alg, group=alg), size=1, show.legend=FALSE) +
  geom_point(data=min.task, aes(x=min.thresh, y=min.stat, color=task, group=task), size=2, show.legend=FALSE) +
  geom_vline(data=min.task, aes(xintercept=min.thresh, color=task, group=task), size=1, show.legend=FALSE) +
  ggtitle("Performance of Downstream Inference Task with Test-Retest Measures") +
  xlab("Threshold") +
  scale_color_discrete(name="") +
  guides(linetype=FALSE) +
  ylab("Statistic")

dMRI Figure

dmri.results <- readRDS('./data/real/knn_dmri_results.rds')
stat.results <- dmri.results$statistics
classifier.results <- dmri.results$problem

max.alg <- data.frame(max.thresh=c(), max.stat=c(), alg=c())
for(un in unique(stat.results$alg)) {
  ss <- stat.results[stat.results$alg == un,]
  ss$stat <- (ss$stat - min(ss$stat))/(max(ss$stat) - min(ss$stat))
  stat.results$stat[stat.results$alg == un] <- ss$stat
  min.pt <- which(ss$stat == max(ss$stat))[1]
  max.alg <- rbind(max.alg, data.frame(max.thresh=ss$thresh[min.pt], max.stat=ss$stat[min.pt], alg=un))
}

class.res <- subset(classifier.results, k == 7 & Metric %in% c("RMSE", "Accuracy") & task != "Lifestyle")
class.res$value[class.res$Metric == "Accuracy"] <- 1 - class.res$value[class.res$Metric=="Accuracy"]
min.task <- data.frame(min.thresh=c(), min.stat=c(), task=c())
for (un in unique(class.res$task)) {
  ss <- class.res[class.res$task == un,]
  ss$value <- (ss$value - min(ss$value))/(max(ss$value) - min(ss$value))
  class.res$value[class.res$task == un] <- ss$value
  min.pt <- which(ss$value == min(ss$value))[1]
  min.task <- rbind(min.task, data.frame(min.thresh=ss$thresh[min.pt], min.stat=ss$value[min.pt], task=un))
}

ggplot(stat.results, aes(x=thresh, color=alg, y=stat, group=alg)) +
  geom_line() +
  theme_bw() +
  geom_line(data=class.res, aes(x=thresh, y=value, color=task, group=task), size=1, linetype=6) +
  geom_point(data=max.alg, aes(x=max.thresh, y=max.stat, color=alg, group=alg), size=2, show.legend=FALSE) +
  geom_vline(data=max.alg, aes(xintercept=max.thresh, color=alg, group=alg), size=1, show.legend=FALSE) +
  geom_point(data=min.task, aes(x=min.thresh, y=min.stat, color=task, group=task), size=2, show.legend=FALSE) +
  geom_vline(data=min.task, aes(xintercept=min.thresh, color=task, group=task), size=1, show.legend=FALSE) +
  ggtitle("(B) dMRI Inference Tasks") +
  xlab("Threshold") +
  scale_color_discrete(name="") +
  guides(linetype=FALSE) +
  ylab("Statistic")