CODEWITHSUNDEEP
r
Dana
Dana

Reputation: 1083

Getting the best variables from best subsets regression in R

I want to run best subset regression on a set of variables and then get the best 3 variables using R. I'm having problems obtaining the best 2 variables. I've included my code below.

set.seed(10)

a <- 1:100
b <- 1:100
c <- 1:100
d <- 1:100
e <- 1:100
f <- 1:100
g <- 1:100
h <- 1:100

data <- data.frame(a, b, c, d, e, f, g, h)

library(leaps)

# best subsets regression
test <- regsubsets(a ~ b + c + d + e + f + g + h, data=data, nbest=4)
# nbest = 4, is the number of subsets of each size that is reported

# plot a table of models showing variables in each model.
summary(test)

# models are ordered by the selection statistic.
plot(test,scale="r2")

#get the variables that are important to the model
coef(test, 2)

#NOTE: THIS DOESN'T GIVE ME THE 2 BEST VARIABLES. IT ONLY GIVES ME THE BEST VARIABLE AT THE 2ND ITERATION. LOOK AT:
coef(test, 1:2)

Your help would be greatly appreciated!

Best, Dana

Upvotes: 1

Views: 2282

Answers (1)

Weihuang Wong
Weihuang Wong

Reputation: 13118

Consider an example with the built-in mtcars dataset:

test <- regsubsets(mpg ~ ., data = mtcars, nbest = 4)

This is the output from summary(test):

summary(test)
# Subset selection object
# Call: regsubsets.formula(mpg ~ ., data = mtcars, nbest = 4)
# 10 Variables  (and intercept)
# <..snip..>
# 4 subsets of each size up to 8
# Selection Algorithm: exhaustive
#          cyl disp hp  drat wt  qsec vs  am  gear carb
# 1  ( 1 ) " " " "  " " " "  "*" " "  " " " " " "  " " 
# 1  ( 2 ) "*" " "  " " " "  " " " "  " " " " " "  " " 
# 1  ( 3 ) " " "*"  " " " "  " " " "  " " " " " "  " " 
# 1  ( 4 ) " " " "  "*" " "  " " " "  " " " " " "  " " 
# 2  ( 1 ) "*" " "  " " " "  "*" " "  " " " " " "  " " 
# 2  ( 2 ) " " " "  "*" " "  "*" " "  " " " " " "  " " 
# 2  ( 3 ) " " " "  " " " "  "*" "*"  " " " " " "  " " 
# 2  ( 4 ) " " " "  " " " "  "*" " "  "*" " " " "  " " 
# 3  ( 1 ) " " " "  " " " "  "*" "*"  " " "*" " "  " " 
# <..snip..>

The sets of coefficients are arranged by the number of independent variables, in subsets of 4 (what we indicated with the nbest argument); hence, coef(test, 1:4) will return coefficients from models with one independent variable, coef(test, 5:8) will be those with two independent variables, and so on. Within each subset, the "best" model comes first. The "best" model with two independent variables will therefore be model 5:

coef(test, 5)
# (Intercept)         cyl          wt 
#   39.686261   -1.507795   -3.190972

Upvotes: 2

Related Questions