R fast way to perturb through data frame - r

Question

I have a data frame that I'm trying to do some scenario analysis with. It looks like this:
Revenue Item_1 Item_2 Item_3
552 200 220 45
1500 400 300 200
2300 600 400 300
I'd like to generate something where 1 item is increased or decreased by some fixed amount (ie 1 unite) like this:
Revenue Item_1 Item_2 Item_3
552 201 220 45
1500 401 300 200
2300 601 400 300
552 200 221 45
1500 400 301 200
2300 600 401 300
552 200 220 46
1500 400 300 201
2300 600 400 301
I'm currently doing it in loop like this but am wondering if there's a faster way:
l1 <- list()
increment_amt <- 1
for(i in c('Item_1','Item_2','Item_3')){
newDf <- df1
newDf[,i] <- newDf[,i] + increment_amt
l1[[i]] <- newDf
}
df2 <- do.call(rbind, l1)
Any suggestions?

Answer 1

With lapply,
do.call(rbind, lapply(names(dat)[2:4], function(x) {dat[,x] <- dat[,x] + 1; dat}))
Revenue Item_1 Item_2 Item_3
1 552 201 220 45
2 1500 401 300 200
3 2300 601 400 300
4 552 200 221 45
5 1500 400 301 200
6 2300 600 401 300
7 552 200 220 46
8 1500 400 300 201
9 2300 600 400 301
Of course, do.call / rbind can be replaced with the data.table's speedier rbindlist, which returns a data.table.
library(data.table)
rbindlist(lapply(names(dat)[2:4], function(x) {dat[,x] <- dat[,x] + 1; dat}))

Answer 2

# Data frame
df <- data.frame(Item_1= c(200, 400, 600),
Item_2= c(220, 300, 400),
Item_3= c(45, 200, 300))
# Perturbation
p <- 1
# Add to all columns
df.new <- apply(diag(ncol(df)) * p, MAR = 1, function(x)data.frame(t(t(df) + x)))
[[1]]
Item_1 Item_2 Item_3
1 201 220 45
2 401 300 200
3 601 400 300
[[2]]
Item_1 Item_2 Item_3
1 200 221 45
2 400 301 200
3 600 401 300
[[3]]
Item_1 Item_2 Item_3
1 200 220 46
2 400 300 201
3 600 400 301

Answer 3

We can write a function and use lapply to achieve this task. df is your original data frame. df_list is a list with all final outputs. You can later use df2 <- do.call(rbind, df_list), or bind_rows from dplyr.
# A function to add 1 to all numbers in a column
add_one <- function(Col, dt){
dt[, Col] <- dt[, Col] + 1
return(dt)
}
# Get the column names
Col_vec <- colnames(df)[2:ncol(df)]
# Apply the add_one function
df_list <- lapply(Col_vec, add_one, dt = df)
# Combine all results
df2 <- dplyr::bind_rows(df_list)

Answer 4

You can use perturb function in R using library(perturb). The code is as follows:
# using the most important features, we create a model
m1 <- lm(revenue ~ item1 + item2 + item3)
#summary(m1)
#anova(m1)
#install.packages("perturb")
library(perturb)
set.seed(1234)
p1_new <- perturb(m1, pvars=c("item1","item2") , prange = c(1,1),niter=20)
p1_new
summary(p1_new)