MATH239: Subset Variable Selection
Variable Selection
Throughout this lab we will use the Hitters dataset. This dataset is from the 1986 and 1987 seasons of the Major League Baseball (MLR).
library(ISLR)
names(Hitters)## [1] "AtBat" "Hits" "HmRun" "Runs" "RBI"
## [6] "Walks" "Years" "CAtBat" "CHits" "CHmRun"
## [11] "CRuns" "CRBI" "CWalks" "League" "Division"
## [16] "PutOuts" "Assists" "Errors" "Salary" "NewLeague"Hitters=na.omit(Hitters)
dim(Hitters)## [1] 263 20sum(is.na(Hitters))## [1] 0We looked at three ways to do variable selection:
- Best Subset Selection
- Forward Selection
- Backward Selection
Best Subset Selection
The best subset algorithm finds the best model for a given number of varibles desired. The leaps package will help us perform variable selection.
library(leaps)
regfit.full=regsubsets(Salary~., Hitters)
summary(regfit.full)## Subset selection object
## Call: regsubsets.formula(Salary ~ ., Hitters)
## 19 Variables (and intercept)
## Forced in Forced out
## AtBat FALSE FALSE
## Hits FALSE FALSE
## HmRun FALSE FALSE
## Runs FALSE FALSE
## RBI FALSE FALSE
## Walks FALSE FALSE
## Years FALSE FALSE
## CAtBat FALSE FALSE
## CHits FALSE FALSE
## CHmRun FALSE FALSE
## CRuns FALSE FALSE
## CRBI FALSE FALSE
## CWalks FALSE FALSE
## LeagueN FALSE FALSE
## DivisionW FALSE FALSE
## PutOuts FALSE FALSE
## Assists FALSE FALSE
## Errors FALSE FALSE
## NewLeagueN FALSE FALSE
## 1 subsets of each size up to 8
## Selection Algorithm: exhaustive
## AtBat Hits HmRun Runs RBI Walks Years CAtBat CHits CHmRun CRuns
## 1 ( 1 ) " " " " " " " " " " " " " " " " " " " " " "
## 2 ( 1 ) " " "*" " " " " " " " " " " " " " " " " " "
## 3 ( 1 ) " " "*" " " " " " " " " " " " " " " " " " "
## 4 ( 1 ) " " "*" " " " " " " " " " " " " " " " " " "
## 5 ( 1 ) "*" "*" " " " " " " " " " " " " " " " " " "
## 6 ( 1 ) "*" "*" " " " " " " "*" " " " " " " " " " "
## 7 ( 1 ) " " "*" " " " " " " "*" " " "*" "*" "*" " "
## 8 ( 1 ) "*" "*" " " " " " " "*" " " " " " " "*" "*"
## CRBI CWalks LeagueN DivisionW PutOuts Assists Errors NewLeagueN
## 1 ( 1 ) "*" " " " " " " " " " " " " " "
## 2 ( 1 ) "*" " " " " " " " " " " " " " "
## 3 ( 1 ) "*" " " " " " " "*" " " " " " "
## 4 ( 1 ) "*" " " " " "*" "*" " " " " " "
## 5 ( 1 ) "*" " " " " "*" "*" " " " " " "
## 6 ( 1 ) "*" " " " " "*" "*" " " " " " "
## 7 ( 1 ) " " " " " " "*" "*" " " " " " "
## 8 ( 1 ) " " "*" " " "*" "*" " " " " " "# Lets check k=0, 1, 2, ..., 19
regfit.full19=regsubsets(Salary~., Hitters, nvmax=19)
reg.summary=summary(regfit.full19)
reg.summary## Subset selection object
## Call: regsubsets.formula(Salary ~ ., Hitters, nvmax = 19)
## 19 Variables (and intercept)
## Forced in Forced out
## AtBat FALSE FALSE
## Hits FALSE FALSE
## HmRun FALSE FALSE
## Runs FALSE FALSE
## RBI FALSE FALSE
## Walks FALSE FALSE
## Years FALSE FALSE
## CAtBat FALSE FALSE
## CHits FALSE FALSE
## CHmRun FALSE FALSE
## CRuns FALSE FALSE
## CRBI FALSE FALSE
## CWalks FALSE FALSE
## LeagueN FALSE FALSE
## DivisionW FALSE FALSE
## PutOuts FALSE FALSE
## Assists FALSE FALSE
## Errors FALSE FALSE
## NewLeagueN FALSE FALSE
## 1 subsets of each size up to 19
## Selection Algorithm: exhaustive
## AtBat Hits HmRun Runs RBI Walks Years CAtBat CHits CHmRun CRuns
## 1 ( 1 ) " " " " " " " " " " " " " " " " " " " " " "
## 2 ( 1 ) " " "*" " " " " " " " " " " " " " " " " " "
## 3 ( 1 ) " " "*" " " " " " " " " " " " " " " " " " "
## 4 ( 1 ) " " "*" " " " " " " " " " " " " " " " " " "
## 5 ( 1 ) "*" "*" " " " " " " " " " " " " " " " " " "
## 6 ( 1 ) "*" "*" " " " " " " "*" " " " " " " " " " "
## 7 ( 1 ) " " "*" " " " " " " "*" " " "*" "*" "*" " "
## 8 ( 1 ) "*" "*" " " " " " " "*" " " " " " " "*" "*"
## 9 ( 1 ) "*" "*" " " " " " " "*" " " "*" " " " " "*"
## 10 ( 1 ) "*" "*" " " " " " " "*" " " "*" " " " " "*"
## 11 ( 1 ) "*" "*" " " " " " " "*" " " "*" " " " " "*"
## 12 ( 1 ) "*" "*" " " "*" " " "*" " " "*" " " " " "*"
## 13 ( 1 ) "*" "*" " " "*" " " "*" " " "*" " " " " "*"
## 14 ( 1 ) "*" "*" "*" "*" " " "*" " " "*" " " " " "*"
## 15 ( 1 ) "*" "*" "*" "*" " " "*" " " "*" "*" " " "*"
## 16 ( 1 ) "*" "*" "*" "*" "*" "*" " " "*" "*" " " "*"
## 17 ( 1 ) "*" "*" "*" "*" "*" "*" " " "*" "*" " " "*"
## 18 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*" "*" " " "*"
## 19 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*" "*" "*" "*"
## CRBI CWalks LeagueN DivisionW PutOuts Assists Errors NewLeagueN
## 1 ( 1 ) "*" " " " " " " " " " " " " " "
## 2 ( 1 ) "*" " " " " " " " " " " " " " "
## 3 ( 1 ) "*" " " " " " " "*" " " " " " "
## 4 ( 1 ) "*" " " " " "*" "*" " " " " " "
## 5 ( 1 ) "*" " " " " "*" "*" " " " " " "
## 6 ( 1 ) "*" " " " " "*" "*" " " " " " "
## 7 ( 1 ) " " " " " " "*" "*" " " " " " "
## 8 ( 1 ) " " "*" " " "*" "*" " " " " " "
## 9 ( 1 ) "*" "*" " " "*" "*" " " " " " "
## 10 ( 1 ) "*" "*" " " "*" "*" "*" " " " "
## 11 ( 1 ) "*" "*" "*" "*" "*" "*" " " " "
## 12 ( 1 ) "*" "*" "*" "*" "*" "*" " " " "
## 13 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 14 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 15 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 16 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 17 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*"
## 18 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*"
## 19 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*"Exercise #1: Pick the best size (Look at plots!)
First, lets take the indirect route and look at the mathematically adjusted statistics (Cp, Adjusted-Rsquared, BIC). Keep in mind that we want to minimize Cp and BIC but maximize the Adjusted-Rsquared.
names(reg.summary)## [1] "which" "rsq" "rss" "adjr2" "cp" "bic" "outmat" "obj"# This first line will create a 2x2 grid of plots in base R
# This redefines the defaults so, we need to make sure we change it back after
par(mfrow=c(2,2))
plot(reg.summary$rss, xlab="Number of Variables", ylab="RSS",
type="l")
plot(reg.summary$adjr2, xlab="Number of Variables",
ylab="Adjusted RSq", type="l")
which.max(reg.summary$adjr2)## [1] 11points(11, reg.summary$adjr2[11], col="red",
cex=2, pch=20)
plot(reg.summary$cp, xlab="Number of Variables",
ylab="Cp", type="l")
which.min(reg.summary$cp)## [1] 10points(10, reg.summary$cp[10], col="red",
cex=2, pch=20)
plot(reg.summary$bic, xlab="Number of Variables",
ylab="BIC", type="l")
which.min(reg.summary$bic)## [1] 6points(6, reg.summary$bic[6], col="red",
cex=2, pch=20)
Foward and Backward Stepwise Selection
Exercise #2: Computational Time Vs. Best Model
We know that the best subset approach is the most computationally expensive, so we consider the alternatives of forward and backward selection. However, in class we have also noted that this may result in suboptimal models.
# FORWARD STEPWISE SELECTION
regfit.fwd=regsubsets(Salary~., data=Hitters, nvmax=19,
method="forward")
summary(regfit.fwd) #Notice that once a variable is in its always in## Subset selection object
## Call: regsubsets.formula(Salary ~ ., data = Hitters, nvmax = 19, method = "forward")
## 19 Variables (and intercept)
## Forced in Forced out
## AtBat FALSE FALSE
## Hits FALSE FALSE
## HmRun FALSE FALSE
## Runs FALSE FALSE
## RBI FALSE FALSE
## Walks FALSE FALSE
## Years FALSE FALSE
## CAtBat FALSE FALSE
## CHits FALSE FALSE
## CHmRun FALSE FALSE
## CRuns FALSE FALSE
## CRBI FALSE FALSE
## CWalks FALSE FALSE
## LeagueN FALSE FALSE
## DivisionW FALSE FALSE
## PutOuts FALSE FALSE
## Assists FALSE FALSE
## Errors FALSE FALSE
## NewLeagueN FALSE FALSE
## 1 subsets of each size up to 19
## Selection Algorithm: forward
## AtBat Hits HmRun Runs RBI Walks Years CAtBat CHits CHmRun CRuns
## 1 ( 1 ) " " " " " " " " " " " " " " " " " " " " " "
## 2 ( 1 ) " " "*" " " " " " " " " " " " " " " " " " "
## 3 ( 1 ) " " "*" " " " " " " " " " " " " " " " " " "
## 4 ( 1 ) " " "*" " " " " " " " " " " " " " " " " " "
## 5 ( 1 ) "*" "*" " " " " " " " " " " " " " " " " " "
## 6 ( 1 ) "*" "*" " " " " " " "*" " " " " " " " " " "
## 7 ( 1 ) "*" "*" " " " " " " "*" " " " " " " " " " "
## 8 ( 1 ) "*" "*" " " " " " " "*" " " " " " " " " "*"
## 9 ( 1 ) "*" "*" " " " " " " "*" " " "*" " " " " "*"
## 10 ( 1 ) "*" "*" " " " " " " "*" " " "*" " " " " "*"
## 11 ( 1 ) "*" "*" " " " " " " "*" " " "*" " " " " "*"
## 12 ( 1 ) "*" "*" " " "*" " " "*" " " "*" " " " " "*"
## 13 ( 1 ) "*" "*" " " "*" " " "*" " " "*" " " " " "*"
## 14 ( 1 ) "*" "*" "*" "*" " " "*" " " "*" " " " " "*"
## 15 ( 1 ) "*" "*" "*" "*" " " "*" " " "*" "*" " " "*"
## 16 ( 1 ) "*" "*" "*" "*" "*" "*" " " "*" "*" " " "*"
## 17 ( 1 ) "*" "*" "*" "*" "*" "*" " " "*" "*" " " "*"
## 18 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*" "*" " " "*"
## 19 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*" "*" "*" "*"
## CRBI CWalks LeagueN DivisionW PutOuts Assists Errors NewLeagueN
## 1 ( 1 ) "*" " " " " " " " " " " " " " "
## 2 ( 1 ) "*" " " " " " " " " " " " " " "
## 3 ( 1 ) "*" " " " " " " "*" " " " " " "
## 4 ( 1 ) "*" " " " " "*" "*" " " " " " "
## 5 ( 1 ) "*" " " " " "*" "*" " " " " " "
## 6 ( 1 ) "*" " " " " "*" "*" " " " " " "
## 7 ( 1 ) "*" "*" " " "*" "*" " " " " " "
## 8 ( 1 ) "*" "*" " " "*" "*" " " " " " "
## 9 ( 1 ) "*" "*" " " "*" "*" " " " " " "
## 10 ( 1 ) "*" "*" " " "*" "*" "*" " " " "
## 11 ( 1 ) "*" "*" "*" "*" "*" "*" " " " "
## 12 ( 1 ) "*" "*" "*" "*" "*" "*" " " " "
## 13 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 14 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 15 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 16 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 17 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*"
## 18 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*"
## 19 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*"# BACKWARD STEPWISE SELECTION
regfit.bwd=regsubsets(Salary~., data=Hitters, nvmax=19,
method="backward")
summary(regfit.bwd)## Subset selection object
## Call: regsubsets.formula(Salary ~ ., data = Hitters, nvmax = 19, method = "backward")
## 19 Variables (and intercept)
## Forced in Forced out
## AtBat FALSE FALSE
## Hits FALSE FALSE
## HmRun FALSE FALSE
## Runs FALSE FALSE
## RBI FALSE FALSE
## Walks FALSE FALSE
## Years FALSE FALSE
## CAtBat FALSE FALSE
## CHits FALSE FALSE
## CHmRun FALSE FALSE
## CRuns FALSE FALSE
## CRBI FALSE FALSE
## CWalks FALSE FALSE
## LeagueN FALSE FALSE
## DivisionW FALSE FALSE
## PutOuts FALSE FALSE
## Assists FALSE FALSE
## Errors FALSE FALSE
## NewLeagueN FALSE FALSE
## 1 subsets of each size up to 19
## Selection Algorithm: backward
## AtBat Hits HmRun Runs RBI Walks Years CAtBat CHits CHmRun CRuns
## 1 ( 1 ) " " " " " " " " " " " " " " " " " " " " "*"
## 2 ( 1 ) " " "*" " " " " " " " " " " " " " " " " "*"
## 3 ( 1 ) " " "*" " " " " " " " " " " " " " " " " "*"
## 4 ( 1 ) "*" "*" " " " " " " " " " " " " " " " " "*"
## 5 ( 1 ) "*" "*" " " " " " " "*" " " " " " " " " "*"
## 6 ( 1 ) "*" "*" " " " " " " "*" " " " " " " " " "*"
## 7 ( 1 ) "*" "*" " " " " " " "*" " " " " " " " " "*"
## 8 ( 1 ) "*" "*" " " " " " " "*" " " " " " " " " "*"
## 9 ( 1 ) "*" "*" " " " " " " "*" " " "*" " " " " "*"
## 10 ( 1 ) "*" "*" " " " " " " "*" " " "*" " " " " "*"
## 11 ( 1 ) "*" "*" " " " " " " "*" " " "*" " " " " "*"
## 12 ( 1 ) "*" "*" " " "*" " " "*" " " "*" " " " " "*"
## 13 ( 1 ) "*" "*" " " "*" " " "*" " " "*" " " " " "*"
## 14 ( 1 ) "*" "*" "*" "*" " " "*" " " "*" " " " " "*"
## 15 ( 1 ) "*" "*" "*" "*" " " "*" " " "*" "*" " " "*"
## 16 ( 1 ) "*" "*" "*" "*" "*" "*" " " "*" "*" " " "*"
## 17 ( 1 ) "*" "*" "*" "*" "*" "*" " " "*" "*" " " "*"
## 18 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*" "*" " " "*"
## 19 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*" "*" "*" "*"
## CRBI CWalks LeagueN DivisionW PutOuts Assists Errors NewLeagueN
## 1 ( 1 ) " " " " " " " " " " " " " " " "
## 2 ( 1 ) " " " " " " " " " " " " " " " "
## 3 ( 1 ) " " " " " " " " "*" " " " " " "
## 4 ( 1 ) " " " " " " " " "*" " " " " " "
## 5 ( 1 ) " " " " " " " " "*" " " " " " "
## 6 ( 1 ) " " " " " " "*" "*" " " " " " "
## 7 ( 1 ) " " "*" " " "*" "*" " " " " " "
## 8 ( 1 ) "*" "*" " " "*" "*" " " " " " "
## 9 ( 1 ) "*" "*" " " "*" "*" " " " " " "
## 10 ( 1 ) "*" "*" " " "*" "*" "*" " " " "
## 11 ( 1 ) "*" "*" "*" "*" "*" "*" " " " "
## 12 ( 1 ) "*" "*" "*" "*" "*" "*" " " " "
## 13 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 14 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 15 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 16 ( 1 ) "*" "*" "*" "*" "*" "*" "*" " "
## 17 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*"
## 18 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*"
## 19 ( 1 ) "*" "*" "*" "*" "*" "*" "*" "*"# DIFFERENT MODELS OF SIZE 7
coef(regfit.full, 7)## (Intercept) Hits Walks CAtBat CHits
## 79.4509472 1.2833513 3.2274264 -0.3752350 1.4957073
## CHmRun DivisionW PutOuts
## 1.4420538 -129.9866432 0.2366813coef(regfit.fwd, 7)## (Intercept) AtBat Hits Walks CRBI
## 109.7873062 -1.9588851 7.4498772 4.9131401 0.8537622
## CWalks DivisionW PutOuts
## -0.3053070 -127.1223928 0.2533404coef(regfit.bwd, 7)## (Intercept) AtBat Hits Walks CRuns
## 105.6487488 -1.9762838 6.7574914 6.0558691 1.1293095
## CWalks DivisionW PutOuts
## -0.7163346 -116.1692169 0.3028847Exercise #3: Cross-Validation
Do this for extra practice at home!