CODEWITHSUNDEEP
rnormalization
vic
vic

Reputation: 400

R - Normalize values (based on some of the values)

I want to normalize data in R but not in a specific range (e.g. 0 to 1). I have a table like the following:

 benchmark  |   technique     | ipc
correlation | no_compression  | 0.5
correlation | compression-bdi | 0.6
  trisolv   | no_compression  | 0.6
  trisolv   | compression-bdi | 0.7

I want the IPC value of no_compression for every benchmark to be 1. The remaining techniques for a specific benchmark will be based on that no_compression value. So for example, the IPC value for compression-bdi for correlation would be 1.2.

Is there any function that I could use? I can only find mentions of normalizing to a certain range.

Upvotes: 0

Views: 144

Answers (2)

Ronak Shah
Ronak Shah

Reputation: 388862

You could also use match which returns the index of first match to find "no_compression" technique 

library(dplyr)

df %>%
   group_by(benchmark) %>%
   mutate(ipc = ipc/ipc[match('no_compression', technique)])

#  benchmark   technique         ipc
#  <fct>       <fct>           <dbl>
#1 correlation no_compression   1   
#2 correlation compression-bdi  1.2 
#3 trisolv     no_compression   1   
#4 trisolv     compression-bdi  1.17

Using data.table that would be

library(data.table)
setDT(df)[, ipc := ipc/ipc[match('no_compression', technique)], benchmark]

Upvotes: 0

Axeman
Axeman

Reputation: 35207

Using dplyr:

df %>% 
  group_by(benchmark) %>% 
  mutate(ipc_standardized = ipc / ipc[technique == 'no_compression'])

# A tibble: 4 x 4
# Groups:   benchmark [2]
  benchmark   technique         ipc ipc_standardized
  <chr>       <chr>           <dbl>            <dbl>
1 correlation no_compression    0.5             1   
2 correlation compression-bdi   0.6             1.2 
3 trisolv     no_compression    0.6             1   
4 trisolv     compression-bdi   0.7             1.17

Or using base R:

df$ipc_standarized <- unlist(lapply(
  split(df, df$benchmark), 
  function(.) .$ipc / .$ipc[.$technique == 'no_compression'])
)

Upvotes: 2

Related Questions