PDFの中に埋め込まれているテーブルからデータを抜き出したいという状況って頻繁にあると思います。まぁもしもそんな状況は一度もなかったとしても、それが簡単にできるということでせっかくなのでちょっと試してみようと思います。Rを使います。
きっかけは、Exploratory西田さんのこのツイートです。tabulizerというライブラリを使うようです。
In R, these 5 lines are all you need to extract table data from 85 pages long PDF. #rstats
library(tabulizer)
library(purer)
library(data.table)pdf <- extract_tables("file.pdf")
df <- map(pdf, function(x){https://t.co/Bz4BWtBUrJ.frame(x)})
rbindlist(df, fill=TRUE)— Kan Nishida 🇺🇸❤️🇯🇵 (@KanAugust) April 20, 2020
この記事も参考になります。
WHOの新型コロナウィルスのレポートからデータを抜き出してみようと思います。

pdfからtableの部分を抜き出し、データフレームで格納するまでは、もうこれだけです。
library(tabulizer) library(purrr) df_combined <- tabulizer::extract_tables("20200424-sitrep-95-covid-19.pdf") %>% purrr::map_dfr(as.data.frame)
すると、データフレームはこんな感じ。
せっかくなんできれいにしていきます。
まずは改行コードやフランス語?を置換。
library(dplyr) library(stringr) df_combined %>% mutate(V1 = str_replace(V1, "\r", " ")) %>% mutate(V1 = str_replace(V1, "ã", "a")) %>% mutate(V1 = str_replace(V1, "í", "i")) %>% mutate(V1 = str_replace(V1, "é", "e")) %>% mutate(V1 = str_replace(V1, "ô", "o")) %>% mutate(V1 = str_replace(V1, "ç", "c")) %>%
ラオスのように国名が改行されてしまっている場合は、念の為1つ前の行を参照しながら修正。ラオス以外も修正。
mutate(V1_prev = lag(V1, n = 1)) %>% # 1つ前の行に入っていた文字列 mutate(V1 = ifelse(V1_prev == "Lao People's" & V1 == "Democratic Republic", "Lao People's Democratic Republic", V1)) %>% mutate(V1 = ifelse(V1_prev == "(Commonwealth of" & V1 == "the)", "Northern Mariana Islands (Commonwealth of the)", V1)) %>% mutate(V1 = ifelse(V1_prev == "United Republic of" & V1 == "Tanzania", "United Republic of Tanzania", V1)) %>% mutate(V1 = ifelse(V1_prev == "Central African" & V1 == "Republic", "Central African Republic", V1)) %>% mutate(V1 = ifelse(V1_prev == "Sao Tome and" & V1 == "Principe", "Sao Tome and Principe", V1)) %>%
簡単な修正は国名の一部を元に、正規表現の置換で対応。
mutate(V1 = gsub("Kosovo.*", "Kosovo", V1)) %>% mutate(V1 = gsub(".*d’Ivoire", "Cote d'Ivoire", V1)) %>% mutate(V1 = gsub(".*conveyance \\(Diamond.*", "International conveyance (Diamond Princess)", V1)) %>%
国名の列だけでなくTransmission classificationの列も修正。
mutate(V6 = str_replace(V6, "\r", " ")) %>% mutate(V6 = ifelse(V6 == "transmission", "Community transmission", V6)) %>%
見出しや総計の不要な行を削除。
filter(V1 != "") %>% filter(V1 != "Grand total") %>% filter(V1 != "Territory/Area †") %>% filter(V2 != "") %>%
作業用に作成した列は不要なので、元々の列のみを選択。
select(c("V1", "V2", "V3", "V4", "V5", "V6", "V7")) %>%
列名を改めて正しく設定。
rename("Reporting Country/Territory/Area"=V1 ,"Total confirmed cases"=V2 ,"Total confirmed new cases"=V3 ,"Total deaths"=V4 ,"Total new deaths"=V5 ,"Transmission classification"=V6 ,"Days since last reported case"=V7)
これでデータフレームはなかなかきれいになり、
最終的にコードはこうなりました。
library(tabulizer) library(purrr) library(dplyr) library(stringr) df_combined <- tabulizer::extract_tables("1_downloadFiles/20200424-sitrep-95-covid-19.pdf") %>% purrr::map_dfr(as.data.frame) df_combined <- df_combined %>% mutate(V1 = str_replace(V1, "\r", " ")) %>% mutate(V1 = str_replace(V1, "ã", "a")) %>% mutate(V1 = str_replace(V1, "í", "i")) %>% mutate(V1 = str_replace(V1, "é", "e")) %>% mutate(V1 = str_replace(V1, "ô", "o")) %>% mutate(V1 = str_replace(V1, "ç", "c")) %>% mutate(V1_prev = lag(V1, n = 1)) %>% mutate(V1 = ifelse(V1_prev == "Lao People's" & V1 == "Democratic Republic", "Lao People's Democratic Republic", V1)) %>% mutate(V1 = ifelse(V1_prev == "(Commonwealth of" & V1 == "the)", "Northern Mariana Islands (Commonwealth of the)", V1)) %>% mutate(V1 = ifelse(V1_prev == "United Republic of" & V1 == "Tanzania", "United Republic of Tanzania", V1)) %>% mutate(V1 = ifelse(V1_prev == "Central African" & V1 == "Republic", "Central African Republic", V1)) %>% mutate(V1 = ifelse(V1_prev == "Sao Tome and" & V1 == "Principe", "Sao Tome and Principe", V1)) %>% mutate(V1 = gsub("Kosovo.*", "Kosovo", V1)) %>% mutate(V1 = gsub(".*d’Ivoire", "Cote d'Ivoire", V1)) %>% mutate(V1 = gsub(".*conveyance \\(Diamond.*", "International conveyance (Diamond Princess)", V1)) %>% mutate(V6 = str_replace(V6, "\r", " ")) %>% mutate(V6 = ifelse(V6 == "transmission", "Community transmission", V6)) %>% filter(V1 != "") %>% filter(V1 != "Grand total") %>% filter(V1 != "Territory/Area †") %>% filter(V2 != "") %>% select(c("V1", "V2", "V3", "V4", "V5", "V6", "V7")) %>% rename("Reporting Country/Territory/Area"=V1 ,"Total confirmed cases"=V2 ,"Total confirmed new cases"=V3 ,"Total deaths"=V4 ,"Total new deaths"=V5 ,"Transmission classification"=V6 ,"Days since last reported case"=V7) write.csv(df_combined, "20200424-sitrep-95-covid-19.csv", row.names = FALSE)
Congoが2つあって、あれWHO間違っている?と思ったが、コンゴってコンゴ共和国とコンゴ民主共和国と2つあるんですね……知らなかったです。