Reads in the output file of a monte carlo simulation for an epsilon-Greedy bandit algorithm implementation

epsilon-Greedy_sanity_check.R

library("dplyr")
library("tidyr")
library("ggplot2")

epsilonGreedySanityCheck <- read.csv("~/Desktop/epsilonGreedySanityCheck.txt", header=FALSE)
names(epsilonGreedySanityCheck) <- c("testRun",
                                     "iterationNum",
                                     "epsilon",
                                     "nArms",
                                     "exploreOrExploit",
                                     "chosenIndex",
                                     "iterationRandomNumber")

epsilonGreedySanityCheck$testRun <- as.factor(epsilonGreedySanityCheck$testRun)
epsilonGreedySanityCheck$iterationNum <- as.factor(epsilonGreedySanityCheck$iterationNum)

check <- epsilonGreedySanityCheck %>%
  group_by(testRun) %>%
  filter( testRun == 0)
  #mutate(randomMoreThanEpsilon = ifelse(iterationRandomNumber > epsilon, as.factor("yes"), as.factor("no")))
  #summarize(cntRandom = count(randomMoreThanEpsilon))

check %>% head()

basegraph <- ggplot(data = epsilonGreedySanityCheck) + theme_bw()

basegraph + geom_histogram(aes(x=iterationRandomNumber, fill=testRun), alpha=0.9) +
  labs(title = "# of times random number was chosen to explore or exploit")

exploreOrExplotPercent <- epsilonGreedySanityCheck %>%
  group_by(testRun, exploreOrExploit) %>%
  summarize(cnt = n()) %>%
  spread(exploreOrExploit, cnt) %>%
  mutate(ratio = (explore / (explore + exploit)))

exploreOrExplotPercent

testRunIterations <- epsilonGreedySanityCheck %>%
  group_by(testRun) %>%
  summarize(meanEps = mean(epsilon),
            maxEps  = max(epsilon)) %>%
  select(testRun, meanEps, maxEps)

testRunIterations

# gut check does the calculated epsilon = epsilon given to the program?
gutCheck <- inner_join(exploreOrExplotPercent, testRunIterations, by="testRun") %>%
  mutate(epsDiff = (meanEps - ratio))

# 7 = epsDiff
diffs <- as.vector(t(gutCheck[, 7]))
diffs
sum(diffs)
mean(diffs)
median(diffs)


# is the chosen index random?
chosenIndexCheck <- epsilonGreedySanityCheck %>%
  filter(exploreOrExploit == "explore") %>%
  group_by(testRun, chosenIndex) %>%
  summarize(idxCnt = n()) %>%
  ungroup()

chosenIndexCheck

ggplot(data=chosenIndexCheck) +
  geom_bar(aes(x=chosenIndex, y=idxCnt), stat="identity") +
  facet_wrap(~ testRun)