Different sampling techniques to generate poems based on Walt Whitman's 'Song of Myself'

song_of_myself_sampling.R

library(dplyr)
library(ggplot2)

## https://www.poetryfoundation.org/poems/45477/song-of-myself-1892-version
## https://docs.google.com/spreadsheets/d/1f7JiBosfa_ralv_Pzx-Q_2IjwIj-HtVupb1V0Cwq1co/edit?usp=sharing

poem <- read.csv("song_of_myself.csv")

#### EDA ####

byStanza <- poem %>%
  group_by(stanza) %>%
  summarise(count = n())

ggplot(byStanza, aes(x = count)) +
  geom_histogram() +
  theme_minimal() +
  xlab("number of lines") +
  ggtitle("Walt Whitman's 'Song of Myself'", subtitle = "Stanza Lengths")

summary(byStanza$count)

prob <- byStanza %>%
  mutate(prop = count / nrow(poem)) %>%
  select(prop) %>%
  as.vector() %>%
  unname() %>%
  unlist()


#### SRS ####

numStanzas <- max(poem$stanza)

set.seed(112514)

## a poem that has the same number of lines as there were number of stanzas in original poem
poem$line[sample(1:nrow(poem), numStanzas, replace = F)]


set.seed(112516)

## a poem that has the same number of lines as there were number of stanzas in original poem
poem$line[sample(1:nrow(poem), numStanzas, replace = F)]



#### stratified ####

set.seed(112517)

## one line per stanza

new_poem <- rep(NA, numStanzas)
for (i in 1:numStanzas) {
  this_stanza <- poem %>% filter(stanza == i)

  new_poem[i] <- this_stanza$line[sample(1:nrow(this_stanza), 1, replace = F)]
}

new_poem

set.seed(112519)

## one line per stanza

new_poem <- rep(NA, numStanzas)
for (i in 1:numStanzas) {
  this_stanza <- poem %>% filter(stanza == i)

  new_poem[i] <- this_stanza$line[sample(1:nrow(this_stanza), 1, replace = F)]
}

new_poem


## lines proportional to stanza length

set.seed(112521)


prop_weights <- ceiling(prob * 100)

longer_poem <- sum(prop_weights)

new_poem <- c()
for (i in 1:numStanzas) {
  this_stanza <- poem %>% filter(stanza == i)

  new_poem <- c(new_poem, this_stanza$line[sample(1:nrow(this_stanza), prop_weights[i], replace = F)])
}

new_poem

set.seed(112525)


prop_weights <- ceiling(prob * 100)

longer_poem <- sum(prop_weights)

new_poem <- c()
for (i in 1:numStanzas) {
  this_stanza <- poem %>% filter(stanza == i)

  new_poem <- c(new_poem, this_stanza$line[sample(1:nrow(this_stanza), prop_weights[i], replace = F)])
}

new_poem


#### clustering ####

set.seed(112527)


numClusters <- 5

poem %>%
  filter(stanza %in% sample(1:numStanzas, numClusters, replace = F)) %>%
  select(line) %>%
  c()


set.seed(112527)


numClusters <- 25

poem %>%
  filter(stanza %in% sample(1:numStanzas, numClusters, replace = F)) %>%
  select(line) %>%
  c()

set.seed(112528)


numClusters <- 25

poem %>%
  filter(stanza %in% sample(1:numStanzas, numClusters, replace = F)) %>%
  select(line) %>%
  c()



#### systematic ####

start <- 50
k <- 50

end_line <- floor(nrow(poem) / k) * 50

poem$line[seq(start, end_line, by = k)]



start <- 1
k <- 50

end_line <- floor(nrow(poem) / k) * 50

poem$line[seq(start, end_line, by = k)]



#### multistage ####

# cluster, then SRS

set.seed(112535)

numClusters <- 5
numLines <- min(byStanza$count)

sampled_stanza <- sample(1:numStanzas, numClusters, replace = F)

new_poem <- c()
for (i in sampled_stanza) {
  this_stanza <- poem %>% filter(stanza == i)
  
  new_poem <- c(new_poem, this_stanza$line[sample(1:nrow(this_stanza), numLines, replace = F)])
}

new_poem

set.seed(112535)

numClusters <- 50
numLines <- min(byStanza$count)

sampled_stanza <- sample(1:numStanzas, numClusters, replace = F)

new_poem <- c()
for (i in sampled_stanza) {
  this_stanza <- poem %>% filter(stanza == i)

  new_poem <- c(new_poem, this_stanza$line[sample(1:nrow(this_stanza), numLines, replace = F)])
}

new_poem

set.seed(112539)

numClusters <- 50
numLines <- min(byStanza$count)

sampled_stanza <- sample(1:numStanzas, numClusters, replace = F)

new_poem <- c()
for (i in sampled_stanza) {
  this_stanza <- poem %>% filter(stanza == i)

  new_poem <- c(new_poem, this_stanza$line[sample(1:nrow(this_stanza), numLines, replace = F)])
}

new_poem