CODEWITHSUNDEEP
rdataframe
Heather Clark
Heather Clark

Reputation: 175

identify and mark duplicate rows in r

I would like to identify and mark duplicate rows based on 2 columns. I would like to make a unique identifier for each duplicate so I know not just that the row is a duplicate, but which row it is a duplicate with. I have a dataframe that looks like below with some duplicate item pairs (on fit and sit) and other pairs that are not duplicated. While the item pairs are duplicated, the information they contain is unique (e.g., one row will have a value in Value1 for 1 row, but not Value2 and Value 3, the second or 'duplicate' row will have numbers for Value2 and Value3 just not Value1)

current dataframe

     value1 value2 value3 fit   sit  
[1,] "1"    NA     NA     "it1" "it2"
[2,] NA     "3"    "2"    "it2" "it1"
[3,] "2"    "3"    "4"    "it3" "it4"
[4,] NA     NA     NA     "it4" "it3"
[5,] "5"    NA     NA     "it5" "it6"
[6,] NA     NA     "2"    "it6" "it5"
[7,] NA     "4"    NA     "it7" "it9"

code to generate example dataframe

value1<-c(1,NA,2,NA,5,NA,NA)
value2<-c(NA,3,3,NA,NA,NA, 4)
value3<-c(NA,2,4,NA,NA,2, NA)
fit<-c("it1","it2","it3","it4", "it5", "it6","it7")
sit<-c("it2","it1","it4","it3", "it6", "it5", "it9")
df.now<-cbind(value1,value2,value3, fit, sit)

what I want is to convert it to a dataframe that looks like this:

desired dataframe

     val1 val2 val3 it1   it2  
[1,] "1"  "3"  "2"  "it1" "it2"
[2,] "2"  "3"  "4"  "it3" "it4"
[3,] "5"  NA   "2"  "it5" "it6"
[4,] NA   "4"  NA   "it7" "it9"

I was thinking of doing the following steps: 1. create new variables using fit and sit with the lowest item and highest items to identify duplicate pairs 2. identify duplicated item pairs 3. use ifelse to select and fill in unique information.

I know how to do steps 1 and 3, but am stuck on step 2. I think what I need to do is not just identify TRUE/FALSE duplicate, but perhaps have a column with a unique identifier for each item pair like this (there are 2 extra rows because of my step 1): 

     value1 value2 value3 fit   sit   lit   hit    dup
[1,] "1"    NA     NA     "it1" "it2" "it1" "it2"   1
[2,] NA     "3"    "2"    "it2" "it1" "it1" "it2"   1
[3,] "2"    "3"    "4"    "it3" "it4" "it3" "it4"   2
[4,] NA     NA     NA     "it4" "it3" "it3" "it4"   2
[5,] "5"    NA     NA     "it5" "it6" "it5" "it6"   3
[6,] NA     NA     "2"    "it6" "it5" "it5" "it6"   3
[7,] NA     "4"    NA     "it7" "it9" "it7" "it9"   NA

I am not sure how to do this.

What I am asking for is either help with step 2 or perhaps there is a better way to solve it than the steps I outlined.

Upvotes: 12

Views: 2023

Answers (6)

chinsoon12
chinsoon12

Reputation: 25225

Another data.table option:

library(data.table)
as.data.table(df.now)[, lapply(.SD, function(x) first(x[!is.na(x)])), 
    .(it1=pmin(fit, sit), it2=pmax(fit, sit)), 
    .SDcols=value1:value3]

output:

   it1 it2 value1 value2 value3
1: it1 it2      1      3      2
2: it3 it4      2      3      4
3: it5 it6      5   <NA>      2
4: it7 it9   <NA>      4   <NA>

Upvotes: 2

Joris C.
Joris C.

Reputation: 6234

This can also be done using tidyr's pivot_longer with values_drop_na = TRUE combined with pivot_wider:

library(tidyverse)

mydf %>%
   mutate(it1 = pmin(fit, sit), it2 = pmax(fit, sit)) %>%
   pivot_longer(cols = starts_with("value"), values_drop_na = TRUE) %>%
   pivot_wider(id_cols = c("it1", "it2"))

#> # A tibble: 4 x 5
#>   it1   it2   value1 value2 value3
#>   <chr> <chr>  <int>  <int>  <int>
#> 1 it1   it2        1      3      2
#> 2 it3   it4        2      3      4
#> 3 it5   it6        5     NA      2
#> 4 it7   it9       NA      4     NA

Data

mydf <- structure(list(value1 = c(1L, NA, 2L, NA, 5L, NA, NA), value2 = c(NA, 
3L, 3L, NA, NA, NA, 4L), value3 = c(NA, 2L, 4L, NA, NA, 2L, NA
), fit = c("it1", "it2", "it3", "it4", "it5", "it6", "it7"), 
sit = c("it2", "it1", "it4", "it3", "it6", "it5", "it9")), class = "data.frame", row.names = c(NA, 
-7L))

Upvotes: 1

akrun
akrun

Reputation: 886938

Using melt/dcast from data.table

library(data.table)
dcast(melt(setDT(df.now)[, c('fit1', 'sit1') := .(pmin(fit, sit), 
    pmax(fit, sit))], measure = patterns("^value"), na.rm = TRUE),
     fit1 + sit1 ~ variable, value.var = 'value')
#   fit1 sit1 value1 value2 value3
#1:  it1  it2      1      3      2
#2:  it3  it4      2      3      4
#3:  it5  it6      5     NA      2
#4:  it7  it9     NA      4     NA

data

df.now <- data.frame(value1,value2,value3, fit, sit, stringsAsFactors = FALSE)

Upvotes: 2

jazzurro
jazzurro

Reputation: 23574

Here is my attempt using data.table. Your data is called mydf. First, I sorted fit and sit for each row and created a new variable, group. Then, for each group, I sorted values in the three value columns (i.e., value1, value2, and value3). Finally, I extracted first row for each group.

library(data.table)

mydt <- setDT(mydf)[, group := paste(sort(.SD), collapse = "_"),
                    .SD = c("fit", "sit"), by = 1:nrow(mydf)][,
                        c("value1", "value2", "value3") := lapply(.SD, sort),
                        .SDcols = value1:value3, by = group][, .SD[1], by = group]

mydt[]

#     group value1 value2 value3 fit sit
#1: it1_it2      1      3      2 it1 it2
#2: it3_it4      2      3      4 it3 it4
#3: it5_it6      5     NA      2 it5 it6
#4: it7_it9     NA      4     NA it7 it9

DATA

mydf <- structure(list(value1 = c(1L, NA, 2L, NA, 5L, NA, NA), value2 = c(NA, 
3L, 3L, NA, NA, NA, 4L), value3 = c(NA, 2L, 4L, NA, NA, 2L, NA
), fit = c("it1", "it2", "it3", "it4", "it5", "it6", "it7"), 
sit = c("it2", "it1", "it4", "it3", "it6", "it5", "it9")), class = "data.frame", row.names = c(NA, 
-7L))

Upvotes: 1

tmfmnk
tmfmnk

Reputation: 39858

One dplyr option could be:

df.now %>%
 group_by(pair = paste(pmax(fit, sit), pmin(fit, sit), sep = "_")) %>%
 summarise_at(vars(starts_with("value")), ~ ifelse(all(is.na(.)), 
                                                   NA,
                                                   first(na.omit(.))))

  pair    value1 value2 value3
  <chr>    <dbl>  <dbl>  <dbl>
1 it2_it1      1      3      2
2 it4_it3      2      3      4
3 it6_it5      5     NA      2
4 it9_it7     NA      4     NA

And if you also need the pairs in individual columns, then with the addition of tidyr you can do:

df.now %>%
 group_by(pair = paste(pmax(fit, sit), pmin(fit, sit), sep = "_")) %>%
 summarise_at(vars(starts_with("value")), ~ ifelse(all(is.na(.)), 
                                                   NA,
                                                   first(na.omit(.)))) %>%
 separate(pair, into = c("fit", "hit"), sep = "_", remove = FALSE)

  pair    fit   hit   value1 value2 value3
  <chr>   <chr> <chr>  <dbl>  <dbl>  <dbl>
1 it2_it1 it2   it1        1      3      2
2 it4_it3 it4   it3        2      3      4
3 it6_it5 it6   it5        5     NA      2
4 it9_it7 it9   it7       NA      4     NA

Upvotes: 6

jay.sf
jay.sf

Reputation: 72593

Use !duplicated() after sorting.

df.now[!duplicated(t(apply(df.now[, c("fit", "sit")], 1, sort))), ]
#       value1 value2 value3 fit   sit  
# [1,] "1"    NA     NA     "it1" "it2"
# [2,] "2"    "3"    "4"    "it3" "it4"
# [3,] "5"    NA     NA     "it5" "it6"
# [4,] NA     "4"    NA     "it7" "it9"

Upvotes: 3

Related Questions