Resampling

errors<-read_csv("data/Resampling.csv")

## Parsed with column specification:
## cols(
##   name = col_character(),
##   lda_error = col_double(),
##   qda_error = col_double(),
##   seed = col_double()
## )

e <- c(errors$lda_error, errors$qda_error)
type <- rep(c("lda","qda"),   c(length(errors$lda_error), length(errors$qda_error)))
errors_tidy<-data.frame(e,type)

ggplot(errors_tidy, aes(x = e, fill = type))+geom_histogram(bins = 15, alpha = .75, position = "identity")+labs(x = "Error", title = "LDA vs QDA Error")+theme_bw()

ggplot(errors_tidy, aes(x = e, fill = type))+geom_density(alpha =.5)+labs(x = "Error", title = "LDA vs QDA Error")+theme_bw()

errors_tidy %>% group_by(type) %>% summarize(mean = mean(e), sd = sd(e))

## `summarise()` ungrouping output (override with `.groups` argument)

## # A tibble: 2 x 3
##   type   mean     sd
##   <fct> <dbl>  <dbl>
## 1 lda   0.204 0.0347
## 2 qda   0.193 0.0310

errors %>% mutate(diff_error = qda_error - lda_error) %>% summarize(mean_diff = mean(diff_error), QDA_better = sum(diff_error < 0), LDA_better = sum(diff_error > 0))

## # A tibble: 1 x 3
##   mean_diff QDA_better LDA_better
##       <dbl>      <int>      <int>
## 1   -0.0113         29         13