Customized normalize by group in R

Asked by At

I have a dataframe that looks like this:

group1<-c(rep(1,12))
group2<-c(rep('Low',6), rep('High',6))
var  <-c(1:6,1:6)
var1  <-c(2:13)
var2  <-c(20:31)  
df1<-data.frame(group1,group2,var,var1,var2)
group1<-c(rep(2,12))
group2<-c(rep('Low',6), rep('High',6))
var  <-c(1:6,1:6)
var1  <-c(2:13)
var2  <-c(20:31)  
df2<-data.frame(group1,group2,var,var1,var2)

df<-rbind(df1,df2)

  group1 group2 var var1 var2
1       1    Low   1    2   20
2       1    Low   2    3   21
3       1    Low   3    4   22
4       1    Low   4    5   23
5       1    Low   5    6   24
6       1    Low   6    7   25
7       1   High   1    8   26
8       1   High   2    9   27
9       1   High   3   10   28
10      1   High   4   11   29
11      1   High   5   12   30
12      1   High   6   13   31
13      2    Low   1    2   20
14      2    Low   2    3   21
15      2    Low   3    4   22
16      2    Low   4    5   23
17      2    Low   5    6   24
18      2    Low   6    7   25
19      2   High   1    8   26
20      2   High   2    9   27
21      2   High   3   10   28
22      2   High   4   11   29
23      2   High   5   12   30
24      2   High   6   13   31

I want to do normalize my columns in the following way. For each combination of group1 and group2, I want to divide var1 and var1 columns with their first element. This allows me to construct a common scale/index across the columns of interest. For example, looking at the combination of group1=1 and group2=low, the relevant elements of var1 should be transformed into 2/2,3/2,4/2,5/2,6/2,7/2 respectively for the combination group1=1 and group2=High should be 8/8,9/8,10/8,11/8,12/8,13/8 and so on.

I want to do the above transformations for both var1 and var2. The expected output should look like this:

   group1 group2 var var1 var2 var1_tra var2_tra
1       1    Low   1    2   20    1.000 1.000000
2       1    Low   2    3   21    1.500 1.050000
3       1    Low   3    4   22    2.000 1.100000
4       1    Low   4    5   23    2.500 1.150000
5       1    Low   5    6   24    3.000 1.200000
6       1    Low   6    7   25    3.500 1.250000
7       1   High   1    8   26    1.000 1.000000
8       1   High   2    9   27    1.125 1.038462
9       1   High   3   10   28    1.250 1.076923
10      1   High   4   11   29    1.375 1.115385
11      1   High   5   12   30    1.500 1.153846
12      1   High   6   13   31    1.625 1.192308
13      2    Low   1    2   20    1.000 1.000000
14      2    Low   2    3   21    1.500 1.050000
15      2    Low   3    4   22    2.000 1.100000
16      2    Low   4    5   23    2.500 1.150000
17      2    Low   5    6   24    3.000 1.200000
18      2    Low   6    7   25    3.500 1.250000
19      2   High   1    8   26    1.000 1.000000
20      2   High   2    9   27    1.125 1.038462
21      2   High   3   10   28    1.250 1.076923
22      2   High   4   11   29    1.375 1.115385
23      2   High   5   12   30    1.500 1.153846
24      2   High   6   13   31    1.625 1.192308

NOTE: Numbers could be anything, usually positive real numbers and because my dataframe is really big, cannot know in advance what could be the element that I want to divide with in order to perform such transformations.

2 Answers

1
akrun On Best Solutions

After grouping by 'group1', 'group2', use mutate_at to do the division of the columns selected by the first value of that column

library(dplyr)
df %>%
   group_by(group1, group2) %>% 
   mutate_at(vars(var1, var2), list(tra = ~ ./first(.)))
# A tibble: 24 x 7
# Groups:   group1, group2 [4]
#   group1 group2   var  var1  var2 var1_tra var2_tra
#    <dbl> <fct>  <int> <int> <int>    <dbl>    <dbl>
# 1      1 Low        1     2    20     1        1   
# 2      1 Low        2     3    21     1.5      1.05
# 3      1 Low        3     4    22     2        1.1 
# 4      1 Low        4     5    23     2.5      1.15
# 5      1 Low        5     6    24     3        1.2 
# 6      1 Low        6     7    25     3.5      1.25
# 7      1 High       1     8    26     1        1   
# 8      1 High       2     9    27     1.12     1.04
# 9      1 High       3    10    28     1.25     1.08
#10      1 High       4    11    29     1.38     1.12
# … with 14 more rows

Or using data.table

nm1 <- c("var1", "var2")
nm2 <- paste0(nm1, "_tra")
library(data.table)
setDT(df)[, (nm2) := lapply(.SD, function(x) x/first(x)), 
              by = .(group1, group2), .SDcols = nm1]
0
OmG On

Also you can use from sqldf likes the following:

result <- sqldf('select df.*, (df.var1 + 0.0) / scale.s_var1 as var1_tra, (df.var2 + 0.0) / scale.s_var2 as var2_tra
          from df join 
                  (select group1, group2, min(var1) as s_var1, min(var2) as s_var2 
                   from df
                   group by group1, group2) as scale 
                 on df.group1 = scale.group1 AND df.group2 = scale.group2 
          ')

In the above code first we find the minimum value for var1 and var2 by each group using the following query:

select group1, group2, min(var1) as s_var1, min(var2) as s_var2 
from df
group by group1, group2

And use that as a nested query and joining with the original data frame df on equality over the value of group1 and group2.