CODEWITHSUNDEEP
rdplyrdata.tablena.approx
Prevost
Prevost

Reputation: 697

Correct use of na.approx in data.table and pipe

I am very new to data.table and would like to try it out to see if it makes my analysis faster. I mainly use knitr to compile .rnw files (which I tend to compile many times per hour so I want it to be as fast as possible).

I have posted a sample below and this is by no means a question of comparison agianst data.table and data.frame. I would like to know if I my code below is what it should be.

I am basically joining two data.tables and then need to linearly approximate using na.approx missing NA values. I used the Introduction to data.table vignette from CRAN and JOINing data in R using data.table from R-Pubs.

The code I am using below results in my best attempt at a data.table method taking a long time (in general too, I only added the other code reference).

Also, if anyone knows if there is a way to pipe in na.approx() into a chain and still have the output as a data.frame that would be appreciated. Note the df_merged = as.data.frame(df_merged) line that I would like to get rid of if possible!

Any input is greatly appreciated thank you!

library(data.table)
library(zoo)
library(dplyr)

dt_function_test = function() {
    set.seed(123)
    # data.table
    dt_random = data.table(vals = runif(1E5, 0, 500))
    dt_na = data.table(vals = c(0, 250, 500),
                       ref1 = c(0.33, 0.45, 0.78),
                       ref2 = c(0.12, 0.79, 1))

    dt_merged = merge(dt_random[],
                      dt_na[],
                      all = TRUE)

    dt_merged = dt_merged[, lapply(.SD,
                                   na.approx),
                          by = vals]
}


df_function_test = function() {
    set.seed(123)
    # data.frame
    df_random = data.frame(vals = runif(1E5, 0, 500))
    df_na = data.frame(vals = c(0, 250, 500),
                       ref1 = c(0.33, 0.45, 0.78),
                       ref2 = c(0.12, 0.79, 1))

    df_merged = full_join(df_random,
                          df_na) %>% 
        na.approx

    df_merged = as.data.frame(df_merged)
}

print(system.time(dt_function_test()))
#  user  system elapsed 
# 11.42    0.00   11.46 

print(system.time(df_function_test()))
# Joining, by = "vals"
#    user  system elapsed  
#    0.05    0.05    0.10

Upvotes: 0

Views: 794

Answers (1)

chinsoon12
chinsoon12

Reputation: 25225

Here is a few possible implementation using data.table that performs zoo::na.approx on the ref* columns (note that a larger dataset has been used also):

library(data.table)
library(zoo)

dt_function_test_0 = function() {
    set.seed(123)
    # data.table
    dt_random = data.table(vals = runif(1e7, 0, 500))
    dt_na = data.table(vals = c(0, 250, 500),
        ref1 = c(0.33, 0.45, 0.78),
        ref2 = c(0.12, 0.79, 1))

    cols <- c("ref1", "ref2")

    ##Version 0
    merge(dt_random, dt_na, all=TRUE)[, lapply(.SD, na.approx)]
}


dt_function_test_1 = function() {
    set.seed(123)
    # data.table
    dt_random = data.table(vals = runif(1e7, 0, 500))
    dt_na = data.table(vals = c(0, 250, 500),
        ref1 = c(0.33, 0.45, 0.78),
        ref2 = c(0.12, 0.79, 1))

    cols <- c("ref1", "ref2")

    ##Version 1: using update by reference
    merge(dt_random, dt_na, all = TRUE)[, 
        (cols) := lapply(.SD, na.approx), .SDcols=cols]
}


dt_function_test_2 = function() {
    set.seed(123)
    # data.table
    dt_random = data.table(vals = runif(1e7, 0, 500))
    dt_na = data.table(vals = c(0, 250, 500),
        ref1 = c(0.33, 0.45, 0.78),
        ref2 = c(0.12, 0.79, 1))

    cols <- c("ref1", "ref2")
    ##Version 2: using set
    dt_merged <- merge(dt_random, dt_na, all = TRUE)
    for (x in cols)
        set(dt_merged, j=x, value=na.approx(dt_merged[[x]]))
    dt_merged
}

timing output:

> system.time(dt_function_test_0())
   user  system elapsed 
   5.44    1.90    6.96 

> system.time(dt_function_test_1())
   user  system elapsed 
   3.55    1.30    4.41 

> system.time(dt_function_test_2())
   user  system elapsed 
   3.78    1.19    4.52

Upvotes: 3

Related Questions