CODEWITHSUNDEEP
rdplyrtidyversepurrrfuzzyjoin
Dmytro Fedoriuk
Dmytro Fedoriuk

Reputation: 351

Comparing each row from one data frame with each row of another one in the tidyverse

I need to compare each row from one dataframe to each row of another one:

             id first_name last_name                         account_nr  amount currency comment
1   wW3A4QgpQQd    Lynnett  Labadini      ES46 2569 1625 6669 5490 4624 9655.56      JPY    Z617
2   LsoPIXEMOo5     Velvet   Ritelli  FR60 4478 1591 96PB SIMI FSTO L13 6992.36      PHP    L841
3   L2wBds77Pw8        orv    matfin LB61 6941 CQYE ONER G5T0 KNDU JU5H 6184.38      CAD    o705
4   ME4O9MKlOzO       ring     hecks        BG28 JYPB 4068 09NB FQ7I 6C 4203.54      IDR    Y548
5   d83N7Viwq8k       judd   Riddick       IL36 2200 2898 6944 4508 084 3619.43      IDR    O762
6   1l96680epEy    Edouard  Kapovski   IS73 1064 6186 1231 6178 3743 49 5291.76      BRL    T397
7   7JwvD23oMzC      Jake Rabinovich           KZ80 759G VOHS JHBY L5TY 6994.26      NGN    Y784
8   ZOcg2uprlN6       vere  gravener      SE39 1416 1830 7878 5026 6805 5281.18      UAH    Z890
9   AUrx3nYR2Ks        Bob     Kelso           VS41 5146 7748 1278 5362 4324.12      USD    W312
10 VrDS+DqRG4S1      Mitch  Mitchell           AT65 6306 7334 7478 1908 4221.59      EUR    T352

another one

            id first_name last_name  amount currency comment recipient
1  xGZx1tNE4oa    Lynnett  Labadini 9655.56      JPY    Z617        72
2  nV7NtxiguPQ     Velvet   Ritelli 6992.36      PHP    L841       175
3  Rto0EHOR17k        Orv    Matfin 6184.38      CAD    O705       412
4  2VDMHTJnxcw       Ring     Hecks 4203.54      IDR    Y548        63
5  VQI7I0EZf1q       Judd   Riddick 3619.43      IDR    O163        39
6  w835JEfmJvZ    Edouard Avramovic 5291.76      BRL    T397       240
7  of2FZZXFKY8      Ferdy Petracchi 6994.26      NGN    Y784       102
8  XgUZFhKowB1       Vere  Gravener 5281.18      IDR    U024       111
9  iGO9advyXP3       Temp McKeevers 7364.49      TND    R404       327
10 5BCiYQVhfxM      Arnie   Ashdown 4221.59      ZAR    N988       262

I want to do it with tidyverse, but anoter way is acceptable too. I don't want to use a loop. There is no matches in IDs. The task is to make kind of a fuzzy-join on first_name, last_name, amount, currency, comment columns. One way that I see is spread each row of the first dataframe nrow time of the another one and use map, but I think it's very memory inefficient.

Upvotes: 1

Views: 120

Answers (1)

Arthur Yip
Arthur Yip

Reputation: 6220

See my solution using fuzzyjoin. It basically does spread each row in left for every row in right because I set a high (10) max_dist but you could lower it if you don't want bad matches anyway. Then it uses group_by and top_n to pick out the best matches for each first_name and last_name in the first data frame.

I added your "mismatch" and "label" criteria (see the first 2 columns). You can tweak the matching function options (right now it compares string distance for the five columns you specified, using a specific stringdist method, Levenshtein).

Also, Bob Kelso shows up twice because the best match is tied between 2 (bad) matches so the algorithm has no way to pick one of equally bad matches. 

library(tidyverse); library(fuzzyjoin)

# Load data
df1 <- tibble::tribble(
             ~id, ~first_name,   ~last_name,                          ~account_nr, ~amount, ~currency, ~comment,
   "wW3A4QgpQQd",   "Lynnett",   "Labadini",      "ES46 2569 1625 6669 5490 4624", 9655.56,     "JPY",   "Z617",
   "LsoPIXEMOo5",    "Velvet",    "Ritelli",  "FR60 4478 1591 96PB SIMI FSTO L13", 6992.36,     "PHP",   "L841",
   "L2wBds77Pw8",       "orv",     "matfin", "LB61 6941 CQYE ONER G5T0 KNDU JU5H", 6184.38,     "CAD",   "o705",
   "ME4O9MKlOzO",      "ring",      "hecks",        "BG28 JYPB 4068 09NB FQ7I 6C", 4203.54,     "IDR",   "Y548",
   "d83N7Viwq8k",      "judd",    "Riddick",       "IL36 2200 2898 6944 4508 084", 3619.43,     "IDR",   "O762",
   "1l96680epEy",   "Edouard",   "Kapovski",   "IS73 1064 6186 1231 6178 3743 49", 5291.76,     "BRL",   "T397",
   "7JwvD23oMzC",      "Jake", "Rabinovich",           "KZ80 759G VOHS JHBY L5TY", 6994.26,     "NGN",   "Y784",
   "ZOcg2uprlN6",      "vere",   "gravener",      "SE39 1416 1830 7878 5026 6805", 5281.18,     "UAH",   "Z890",
   "AUrx3nYR2Ks",       "Bob",      "Kelso",           "VS41 5146 7748 1278 5362", 4324.12,     "USD",   "W312",
  "VrDS+DqRG4S1",     "Mitch",   "Mitchell",           "AT65 6306 7334 7478 1908", 4221.59,     "EUR",   "T352"
  )
df2 <- tibble::tribble(
                   ~id, ~first_name,  ~last_name, ~amount, ~currency, ~comment, ~recipient,
         "xGZx1tNE4oa",   "Lynnett",  "Labadini", 9655.56,     "JPY",   "Z617",         72,
         "nV7NtxiguPQ",    "Velvet",   "Ritelli", 6992.36,     "PHP",   "L841",        175,
         "Rto0EHOR17k",       "Orv",    "Matfin", 6184.38,     "CAD",   "O705",        412,
         "2VDMHTJnxcw",      "Ring",     "Hecks", 4203.54,     "IDR",   "Y548",         63,
         "VQI7I0EZf1q",      "Judd",   "Riddick", 3619.43,     "IDR",   "O163",         39,
         "w835JEfmJvZ",   "Edouard", "Avramovic", 5291.76,     "BRL",   "T397",        240,
         "of2FZZXFKY8",     "Ferdy", "Petracchi", 6994.26,     "NGN",   "Y784",        102,
         "XgUZFhKowB1",      "Vere",  "Gravener", 5281.18,     "IDR",   "U024",        111,
         "iGO9advyXP3",      "Temp", "McKeevers", 7364.49,     "TND",   "R404",        327,
         "5BCiYQVhfxM",     "Arnie",   "Ashdown", 4221.59,     "ZAR",   "N988",        262
         )

# Solution using fuzzyjoin
stringdist_left_join(df1, df2, by = c("first_name", "last_name", "amount", "currency", "comment"), 
                     max_dist = 10, ignore_case = TRUE, method = "lv", distance_col = "dist") %>%
  mutate(total.dist = first_name.dist + last_name.dist + amount.dist + currency.dist + comment.dist) %>%
  group_by(first_name.x, last_name.x) %>%
  top_n(-1, total.dist) %>%
  mutate(mismatch = (first_name.dist>0) + (last_name.dist>0) + (amount.dist>0) + (currency.dist>0) + (comment.dist>0),
         label = case_when(mismatch == 0 ~ "match",
                           mismatch == 1 ~ "high",
                           mismatch == 2 ~ "proposed",
                           mismatch > 2 ~ "none",
                           TRUE ~ "")) %>%
  select(label, mismatch, total.dist, everything())
#> # A tibble: 11 x 22
#> # Groups:   first_name.x, last_name.x [10]
#>    label mismatch total.dist id.x  first_name.x last_name.x account_nr
#>    <chr>    <int>      <dbl> <chr> <chr>        <chr>       <chr>     
#>  1 match        0          0 wW3A~ Lynnett      Labadini    ES46 2569~
#>  2 match        0          0 LsoP~ Velvet       Ritelli     FR60 4478~
#>  3 match        0          0 L2wB~ orv          matfin      LB61 6941~
#>  4 match        0          0 ME4O~ ring         hecks       BG28 JYPB~
#>  5 high         1          2 d83N~ judd         Riddick     IL36 2200~
#>  6 high         1          7 1l96~ Edouard      Kapovski    IS73 1064~
#>  7 prop~        2         14 7Jwv~ Jake         Rabinovich  KZ80 759G~
#>  8 prop~        2          7 ZOcg~ vere         gravener    SE39 1416~
#>  9 none         5         20 AUrx~ Bob          Kelso       VS41 5146~
#> 10 none         5         20 AUrx~ Bob          Kelso       VS41 5146~
#> 11 none         4         19 VrDS~ Mitch        Mitchell    AT65 6306~
#> # ... with 15 more variables: amount.x <dbl>, currency.x <chr>,
#> #   comment.x <chr>, id.y <chr>, first_name.y <chr>, last_name.y <chr>,
#> #   amount.y <dbl>, currency.y <chr>, comment.y <chr>, recipient <dbl>,
#> #   amount.dist <dbl>, comment.dist <dbl>, currency.dist <dbl>,
#> #   first_name.dist <dbl>, last_name.dist <dbl>

Created on 2019-03-17 by the reprex package (v0.2.1)

Upvotes: 1

Related Questions