wordle_fast.R

# This function returns the result (black, green, or yellow for each letter) 
# given a guess and an answer 
get_result <- function(g, a) {
  r <- "bbbbb"
  for (i in 1:5) {
    if (substr(g, i, i) == substr(a, i, i)) {
      substr(r, i, i) <- "g"
    }
  }
  
  unique_chars_g <- unique(strsplit(g, "")[[1]])
  for(i in 1:length(unique_chars_g)) {
    occurances_in_a <- unlist(gregexpr(unique_chars_g[i], a))
    occurances_in_g <- unlist(gregexpr(unique_chars_g[i], g))
    remove_g <- occurances_in_g %in% occurances_in_a
    remove_a <- occurances_in_a %in% occurances_in_g
    g_new <- occurances_in_g[!remove_g]
    a_new <- occurances_in_a[!remove_a]
    if (length(a_new) != 0 && length(g_new) != 0) {
      if (a_new[1] != -1) {
        for (j in 1:length(a_new)) {
          substr(r, g_new[j], g_new[j]) <- "y"
          if (length(g_new) == j) {
            break
          }
        }
      }
    }
  }
  return(r)
}

# This function removes all words that wouldn't return the same result that 
# the provided guess returned
subset_data <- function(g, r, mydata) {
  rs <- as.vector(sapply(mydata[,1], get_result, g = g))
  mydata <- mydata[rs == r,]
  return(mydata)
}

# This function is the main solver. Given a subsetted dataset, it returns a 
# vector of "entropies" for each possible guess. The entropy is a measure
# of how well the guess partitions the sample space.
solver <- function(mydata) {
  entropy <- c()
  for (i in 1:nrow(mydata)) {
    entropy[i] <- 0
    guess_test <- mydata[i,1]
    for (j in 1:nrow(mydata)) {
      if (i != j) {
        answer_test <- mydata[j,1]
        data_test <- subset_data(guess_test, get_result(guess_test, answer_test), mydata)
        entropy[i] <- entropy[i] + nrow(data_test)
      }
    }
  }
  return(entropy)
}

# These are all the possible answers for wordle
words <- read.table("possible_words.txt")
for (i in 1:5) {
  words[,i+1] <- substr(words[,1], i, i)
}

# My first guess is usually share
first_guess <- "share"

# This while loop will play through the rounds until there is only one possibility
# or we get it right by giving "ggggg" as the result
while(TRUE) {
  first_guess <- readline(prompt="Guess: ")
  result <- readline(prompt="Result: ")
  if (result == "ggggg") {
    break
  }
  words <- subset_data(first_guess, result, words)
  
  # I've already simulated the best second guess for some possibilities
  # that take a long time (> 5 min) to estimate
  
  if (first_guess == "share" & result == "bbbbb") {
    first_guess <- "unlit"
  } else if (first_guess == "share" & result == "bybbb") {
    first_guess <- "touch"
  } else if (first_guess == "share" & result == "bbybb") {
    first_guess <- "talon"
  } else if (first_guess == "share" & result == "bbbyb") {
    first_guess <- "droit"
  } else if (first_guess == "share" & result == "bbbby") {
    first_guess <- "olden"
  } else if (first_guess == "share" & result == "gbbbb") {
    first_guess <- "stink"
  } else if (first_guess == "share" & result == "bbbbg") {
    first_guess <- "guile"
  } else if (first_guess == "share" & result == "bbbyy") {
    first_guess <- "rider"
  } else if (first_guess == "share" & result == "bbyyy") {
    first_guess <- "alter"
  } else if (first_guess == "share" & result == "bbyyb") {
    first_guess <- "carol"
  } else if (first_guess == "share" & result == "bbyby") {
    first_guess <- "penal"
  } else if (first_guess == "share" & result == "bbbyg") {
    first_guess <- "trope"
  } else {
    entropy <- solver(words)
    first_guess <- words[,1][entropy == min(entropy)]
  }
  
  print(paste("Guess",first_guess))
  if (nrow(words) == 1) {
    break
  }
}