extract_values_from_bar_charts.R

needs(tidyverse, magrittr, png)

# Create a folder for storing the charts
dir.create("BFB_images")

# Loop though constituency codes in England and Wales, downloading the chart for each on from BfB
for(pcon in c(B4B_MRP$westminster_constituency[B4B_MRP$region != "Scotland"])){
  tryCatch(download.file(paste0("https://www.getvoting.org/charts/",pcon,"_1.png"), destfile = paste0("BFB_images/", pcon, ".png")), error = function(e){message("No chart")})
}

# Function for inferring each party’s projected vote share from the chart
process_chart <- function(constituency_code){
  
  # Read in the chart as a matrix of red, green and blue colour values
  img <- readPNG(paste0("BFB_images/",constituency_code,".png"))
  
  # Multiply each colour band by 255 to turn the 0-1 values into standard RGB integer values
  r <- img[,,1] * 255
  g <- img[,,2] * 255
  b <- img[,,3] * 255
    
  # Find the pixel coordinate associated with a vote share of zero by identifying the left-most pixel of the Labour vote share bar
  zero <- c(1:330) %>%
    map_dbl(~{
      min(which(
        # The line below finds pixels that are rgb(213,0,0), the colour of the Labour bars
        r[.x,1:615]==213 & g[.x,1:615]==0 & b[.x,1:615]==0
      )) %>% replace(!is.finite(.), 999)
    }) %>%
    reduce(min)
  
  # Find the pixel value associated with Labour’s projected vote share by identifying the right-most pixel of the Labour vote share bar
  lab <- c(1:330) %>%
    map_dbl(~{
      max(which(
        r[.x,1:615]==213 & g[.x,1:615]==0 & b[.x,1:615]==0
      )) %>% replace(!is.finite(.), 0)
    }) %>%
    reduce(max) %>%
    # Then subtract the pixel number associated with zero, to find the number of horizontal pixels used for Labour’s projected vote share
    subtract(zero)
  
  # Repeat the above step for all other parties
  con <- c(1:330) %>%
    map_dbl(~{
      max(which(
        r[.x,1:615]==0 & g[.x,1:615]==155 & b[.x,1:615]==227
      )) %>% replace(!is.finite(.), 0)
    }) %>%
    reduce(max) %>%
    subtract(zero)
  
  lib <- c(1:330) %>%
    map_dbl(~{
      max(which(
        r[.x,1:615]==253 & g[.x,1:615]==187 & b[.x,1:615]==48
      )) %>% replace(!is.finite(.), 0)
    }) %>%
    reduce(max) %>%
    subtract(zero)
  
  bxp <- c(1:330) %>%
    map_dbl(~{
      max(which(
        r[.x,1:615]==18 & g[.x,1:615]==182 & b[.x,1:615]==207
      )) %>% replace(!is.finite(.), 0)
    }) %>%
    reduce(max) %>%
    subtract(zero)
  
  grn <- c(1:330) %>%
    map_dbl(~{
      max(which(
        r[.x,1:615]==0 & g[.x,1:615]==128 & b[.x,1:615]==102
      )) %>% replace(!is.finite(.), 0)
    }) %>%
    reduce(max) %>%
    subtract(zero)
  
  pc <- c(1:330) %>%
    map_dbl(~{
      max(which(
        r[.x,1:615]==64 & g[.x,1:615]==132 & b[.x,1:615]==40
      )) %>% replace(!is.finite(.), 0)
    }) %>%
    reduce(max) %>%
    subtract(zero)
  
  oth <- c(1:330) %>%
    map_dbl(~{
      max(which(
        r[.x,1:615]==191 & g[.x,1:615]==191 & b[.x,1:615]==191
      )) %>% replace(!is.finite(.), 0)
    }) %>%
    reduce(max) %>%
    subtract(zero)
  
  # If there was no value for Plaid Cymru, the method I’ve used will have returned a negative number (zero minus the pixel associated with zero), so change this to zero
  if(pc < 0){
    pc <- 0
  }
  
  # Now return a little table of party names, vote shares and the constituency code
  tibble(
    party = c("con", "lab", "lib", "bxp", "pc", "grn", "oth"),
    share = c(con, lab, lib, bxp, pc, grn, oth)
  ) %>%
    transmute(constituency_code, party, share = share/sum(share)*100)
  
}

# Loop through all downloaded images, and process each chart, finally returning a big table of all projected vote shares for all parties in all seats
inferred_shares <- list.files(path="BFB_images/") %>%
  str_replace("(\\.png)", "") %>%
  imap_dfr(~{
    message(.y)
    return(process_chart(.x))
  }) %>%
  spread(party, share, 2:3) %>%
  left_join(B4B_MRP %>% dplyr::select(constituency, constituency_code=westminster_constituency)) %>%
  dplyr::select(1,9, everything())