Get rid of nested (unnecessary?) for-loops

user_8852035 0
Viewed 41

I am experienced in Fortran, but quite new in R. In Fortran I am used to nest several do-loops, but I guess there are better methods in R. Some other questions were answered by applying apply, but I am not sure whether this is the right way for me.

I want to do a bias correction for my model data. I know that packages exist for that, but I'd prefer to code it by myself. I have two data.frames, the first contains my model data:

library(dplyr)
x <- round(runif(34698,0,20), 2)
df_a <- data.frame(date=as.Date(0:34697, origin="2006-01-01"),x)
df_a <- setNames(df_a, c("date","daily"))
df_a <- separate(df_a, date, into = c("year", "month", "day"), sep="-")

The second data frame contains the observed and modeled historical monthly means:

df_b <- data.frame(month=seq(01,12,by=1),obs=seq(1.1,12.1,by=1),model=seq(2.2,13.2,by=1))
df_b$month <- ifelse(nchar(df_b$month)!=2,paste0("0",df_b$month),df_b$month)

With the following code, I correct the data of my first data.frame by using the means of each month of the second data.frame. The code works fine, but I think it's not the R-style of coding it. Especially, I would need even more for-loops because I have several model outputs and for each model I have two different scenarios.

system.time(
  for(i in 1:12){
    for (j in 1:nrow(df_a)) {
      if(df_b$month[i]==df_a$month[j]){
        df_a$daily[j] <- df_a$daily[j]+(df_b$obs[i]-df_b$model[i])
      }
    }
  }
)

I would really appreciate anyone how could show me how to "improve" my style of coding in R.

r for-loop
1 Answers

A better option would be to do a left_join and mutate to create the new column

library(dplyr)
df_a1 <- df_a %>% 
            left_join(df_b) %>% 
            mutate(daily = daily + obs + model)

Benchmarks

system.time(df_a %>% 
              left_join(df_b) %>%
              mutate(daily = daily + obs + model))  
#   user  system elapsed 
#  0.201   0.011   0.213

Also, as @parfait mentioned in the comments, a base R version with merge would be

system.time( within(merge(df_a, df_b, by="month", all.x=TRUE), {
              daily <- daily + obs + model}))
#   user  system elapsed 
#  0.260   0.015   0.275

Or with data.table

library(data.table)
system.time(setDT(df_a)[df_b, daily := daily + obs + model, on = .(month)])
#   user  system elapsed 
#  0.198   0.011   0.208

and the OP's for loop 

system.time(
   for(i in 1:12){
     for (j in 1:nrow(df_a)) {
       if(df_b$month[i]==df_a$month[j]){
         df_a$daily[j] <- df_a$daily[j]+(df_b$obs[i]-df_b$model[i])
       }
     }
   }
 )
#   user  system elapsed 
#  9.661   2.741  12.306
user_37322710
Related