We have seen that we can fit an SVM with a non-linear kernel in order to perform classification using a non-linear decision boundary. We will now see that we can also obtain a non-linear decision boundary by performing logistic regression using non-linear transformations of the features.
Generate a data set with n = 500 and p = 2, such that the observations belong to two classes with a quadratic decision boundary between them.
Plot the observations, colored according to their class labels. Your plot should display X1 on the x-axis, and X2 on the yaxis.
set.seed(2)
x1 <- runif(500) - 0.5
x2 <- runif(500) - 0.5
y <- 1 * (x1^2 - x2^2 > 0)
plot(x1[y == 0], x2[y == 0], col = "blue", xlab = "X1", ylab = "X2", pch = "+")
points(x1[y == 1], x2[y == 1], col = "darkgrey", pch = 4)
(c) Fit a logistic regression model to the data, using X1 and X2 as
predictors.
lm.fit <- glm(y ~ x1 + x2, family = binomial)
summary(lm.fit)
##
## Call:
## glm(formula = y ~ x1 + x2, family = binomial)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.271 -1.193 1.097 1.147 1.209
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.07138 0.08959 0.797 0.426
## x1 -0.03532 0.29825 -0.118 0.906
## x2 0.27548 0.30762 0.896 0.370
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 692.50 on 499 degrees of freedom
## Residual deviance: 691.67 on 497 degrees of freedom
## AIC: 697.67
##
## Number of Fisher Scoring iterations: 3
data <- data.frame(x1 = x1, x2 = x2, y = y)
lm.prob <- predict(lm.fit, data, type = "response")
lm.pred <- ifelse(lm.prob > 0.52, 1, 0)
data.pos <- data[lm.pred == 1, ]
data.neg <- data[lm.pred == 0, ]
plot(data.pos$x1, data.pos$x2, col = "blue", xlab = "X1", ylab = "X2", pch = "+")
points(data.neg$x1, data.neg$x2, col = "darkgrey", pch = 4)
(e) Now fit a logistic regression model to the data using non-linear
functions of X1 and X2 as predictors (e.g. X2 1 , X1×X2, log(X2), and so
forth).
lm.fit <- glm(y ~ poly(x1, 2) + poly(x2, 2) + I(x1 * x2), data = data, family = binomial)
## Warning: glm.fit: algorithm did not converge
## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred
summary(lm.fit)
##
## Call:
## glm(formula = y ~ poly(x1, 2) + poly(x2, 2) + I(x1 * x2), family = binomial,
## data = data)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.441e-03 -2.000e-08 2.000e-08 2.000e-08 1.455e-03
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 89.95 48695.53 0.002 0.999
## poly(x1, 2)1 203.19 1143572.92 0.000 1.000
## poly(x1, 2)2 32729.10 895641.74 0.037 0.971
## poly(x2, 2)1 540.94 1130136.06 0.000 1.000
## poly(x2, 2)2 -33281.76 1338007.08 -0.025 0.980
## I(x1 * x2) 198.04 578657.79 0.000 1.000
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 6.9250e+02 on 499 degrees of freedom
## Residual deviance: 4.9969e-06 on 494 degrees of freedom
## AIC: 12
##
## Number of Fisher Scoring iterations: 25
lm.prob <- predict(lm.fit, data, type = "response")
lm.pred <- ifelse(lm.prob > 0.5, 1, 0)
data.pos <- data[lm.pred == 1, ]
data.neg <- data[lm.pred == 0, ]
plot(data.pos$x1, data.pos$x2, col = "blue", xlab = "X1", ylab = "X2", pch = "+")
points(data.neg$x1, data.neg$x2, col = "darkgrey", pch = 4)
(g) Fit a support vector classifier to the data with X1 and X2 as
predictors. Obtain a class prediction for each training observation.
Plot the observations, colored according to the predicted class
labels.
svm.fit <- svm(as.factor(y) ~ x1 + x2, data, kernel = "linear", cost = 0.1)
svm.pred <- predict(svm.fit, data)
data.pos <- data[svm.pred == 1, ]
data.neg <- data[svm.pred == 0, ]
plot(data.pos$x1, data.pos$x2, col = "blue", xlab = "X1", ylab = "X2", pch = "+")
points(data.neg$x1, data.neg$x2, col = "darkgrey", pch = 4)
(h) Fit a SVM using a non-linear kernel to the data. Obtain a class
prediction for each training observation. Plot the observations, colored
according to the predicted class labels.
svm.fit <- svm(as.factor(y) ~ x1 + x2, data, gamma = 1)
svm.pred <- predict(svm.fit, data)
data.pos <- data[svm.pred == 1, ]
data.neg <- data[svm.pred == 0, ]
plot(data.pos$x1, data.pos$x2, col = "blue", xlab = "X1", ylab = "X2", pch = 5)
points(data.neg$x1, data.neg$x2, col = "darkgrey", pch = 5)
(i) Comment on your results.
Using the logistic regression mehtod or the linear kernel, the results are poor in terms of finding the non-linear decision boundaries.If wanting to find non-linear decision boundaries the original logistic regression and the SVM with non-linear kernel were both good models.
In this problem, you will use support vector approaches in order to predict whether a given car gets high or low gas mileage based on the Auto data set.
Create a binary variable that takes on a 1 for cars with gas mileage above the median, and a 0 for cars with gas mileage below the median.
Fit a support vector classifier to the data with various values of cost, in order to predict whether a car gets high or low gas mileage. Report the cross-validation errors associated with different values of this parameter. Comment on your results. Note you will need to fit the classifier without the gas mileage variable to produce sensible results.
##
## Parameter tuning of 'svm':
##
## - sampling method: 10-fold cross validation
##
## - best parameters:
## cost
## 1
##
## - best performance: 0.01025641
##
## - Detailed performance results:
## cost error dispersion
## 1 1e-02 0.07653846 0.03617137
## 2 1e-01 0.04596154 0.03378238
## 3 1e+00 0.01025641 0.01792836
## 4 5e+00 0.02051282 0.02648194
## 5 1e+01 0.02051282 0.02648194
## 6 1e+02 0.03076923 0.03151981
This model uses a 10 fold cross validation. Best parameters are at a cost of 1 and has a performance of 0.0102.
set.seed(21)
tune.out <- tune(svm, mpglevel ~ ., data = Auto, kernel = "polynomial", ranges = list(cost = c(0.1,
1, 5, 10), degree = c(2, 3, 4)))
summary(tune.out)
##
## Parameter tuning of 'svm':
##
## - sampling method: 10-fold cross validation
##
## - best parameters:
## cost degree
## 10 2
##
## - best performance: 0.5435897
##
## - Detailed performance results:
## cost degree error dispersion
## 1 0.1 2 0.5587821 0.04538579
## 2 1.0 2 0.5587821 0.04538579
## 3 5.0 2 0.5587821 0.04538579
## 4 10.0 2 0.5435897 0.05611162
## 5 0.1 3 0.5587821 0.04538579
## 6 1.0 3 0.5587821 0.04538579
## 7 5.0 3 0.5587821 0.04538579
## 8 10.0 3 0.5587821 0.04538579
## 9 0.1 4 0.5587821 0.04538579
## 10 1.0 4 0.5587821 0.04538579
## 11 5.0 4 0.5587821 0.04538579
## 12 10.0 4 0.5587821 0.04538579
The best performance is 0.5435897 that has the lowest cross validation error obtained for a degree of 2 and a cost of 100.
linear.svm <- svm(mpglevel ~ ., data = Auto, kernel = "linear", cost = 1)
poly.svm <- svm(mpglevel ~ ., data = Auto, kernel = "polynomial", cost = 10,
degree = 2)
radial.svm = svm(mpglevel ~ ., data = Auto, kernel = "radial", cost = 10, gamma = 0.01)
plotpairs <- function(fit) {
for (name in names(Auto)[!(names(Auto) %in% c("mpg", "mpglevel", "name"))]) {
plot(fit, Auto, as.formula(paste("mpg~", name, sep = "")))
}
}
plotpairs(linear.svm)
This problem involves the OJ data set which is part of the ISLR2 package.
Create a training set containing a random sample of 800 observations, and a test set containing the remaining observations.
Fit a support vector classifier to the training data using cost = 0.01, with Purchase as the response and the other variables as predictors. Use the summary() function to produce summary statistics, and describe the results obtained.
set.seed(1)
train <- sample(dim(OJ)[1], 800)
train.oj <- OJ[train, ]
test.oj <- OJ[-train, ]
linear.svm <- svm(Purchase ~ ., data = train.oj, kernel = "linear", cost = 0.01)
summary(linear.svm)
##
## Call:
## svm(formula = Purchase ~ ., data = train.oj, kernel = "linear", cost = 0.01)
##
##
## Parameters:
## SVM-Type: C-classification
## SVM-Kernel: linear
## cost: 0.01
##
## Number of Support Vectors: 435
##
## ( 219 216 )
##
##
## Number of Classes: 2
##
## Levels:
## CH MM
435 support vectors are created, where 219 are level Ch and 216 are level MM.
train.pred <- predict(linear.svm, train.oj)
table(train.oj$Purchase, train.pred)
## train.pred
## CH MM
## CH 420 65
## MM 75 240
(65+75)/800
## [1] 0.175
test.pred <- predict(linear.svm, test.oj)
table(test.oj$Purchase, test.pred)
## test.pred
## CH MM
## CH 153 15
## MM 33 69
(15+33)/270
## [1] 0.1777778
The training error is 0.175 and the test error is 0.178.
set.seed(12)
tune.out <- tune(svm, Purchase ~ ., data = train.oj, kernel = "linear", ranges = list(cost = 10^seq(-2, 1, by = 0.25)))
summary(tune.out)
##
## Parameter tuning of 'svm':
##
## - sampling method: 10-fold cross validation
##
## - best parameters:
## cost
## 10
##
## - best performance: 0.17125
##
## - Detailed performance results:
## cost error dispersion
## 1 0.01000000 0.17500 0.03118048
## 2 0.01778279 0.17500 0.02825971
## 3 0.03162278 0.17875 0.02638523
## 4 0.05623413 0.17500 0.03061862
## 5 0.10000000 0.17625 0.03408018
## 6 0.17782794 0.17125 0.02829041
## 7 0.31622777 0.17500 0.03061862
## 8 0.56234133 0.17500 0.03061862
## 9 1.00000000 0.17625 0.03087272
## 10 1.77827941 0.17375 0.03143004
## 11 3.16227766 0.17500 0.03280837
## 12 5.62341325 0.17250 0.03425801
## 13 10.00000000 0.17125 0.03175973
There is an optimal cost of 10 with a best performance of 0.17125
linear.svm <- svm(Purchase ~ ., kernel = "linear", data = train.oj, cost = tune.out$best.parameter$cost)
train.pred <- predict(linear.svm, train.oj)
table(train.oj$Purchase, train.pred)
## train.pred
## CH MM
## CH 423 62
## MM 69 246
(62+69)/800
## [1] 0.16375
test.pred <- predict(linear.svm, test.oj)
table(test.oj$Purchase, test.pred)
## test.pred
## CH MM
## CH 156 12
## MM 28 74
(12+28)/270
## [1] 0.1481481
The tuning of svm caused a decrease in the training error rate to 0.164 and decrease in the test error rate to 0.148.
radial.svm <- svm(Purchase ~ ., kernel = "radial", data = train.oj)
summary(radial.svm)
##
## Call:
## svm(formula = Purchase ~ ., data = train.oj, kernel = "radial")
##
##
## Parameters:
## SVM-Type: C-classification
## SVM-Kernel: radial
## cost: 1
##
## Number of Support Vectors: 373
##
## ( 188 185 )
##
##
## Number of Classes: 2
##
## Levels:
## CH MM
373 support vectors, where 188 vectors are level CH and 185 are level MM.
train.pred <- predict(radial.svm, train.oj)
table(train.oj$Purchase, train.pred)
## train.pred
## CH MM
## CH 441 44
## MM 77 238
(44+77)/800
## [1] 0.15125
test.pred <- predict(radial.svm, test.oj)
table(test.oj$Purchase, test.pred)
## test.pred
## CH MM
## CH 151 17
## MM 33 69
(17+33)/270
## [1] 0.1851852
The training error rate is 0.151 and the test errror rate is 0.185.
set.seed(12)
tune.out <- tune(svm, Purchase ~ ., data = train.oj, kernel = "radial", ranges = list(cost = 10^seq(-2,
1, by = 0.25)))
summary(tune.out)
##
## Parameter tuning of 'svm':
##
## - sampling method: 10-fold cross validation
##
## - best parameters:
## cost
## 0.3162278
##
## - best performance: 0.17
##
## - Detailed performance results:
## cost error dispersion
## 1 0.01000000 0.39375 0.05597929
## 2 0.01778279 0.39375 0.05597929
## 3 0.03162278 0.35000 0.07264832
## 4 0.05623413 0.19875 0.04656611
## 5 0.10000000 0.18750 0.04124790
## 6 0.17782794 0.18000 0.04257347
## 7 0.31622777 0.17000 0.03641962
## 8 0.56234133 0.17375 0.04619178
## 9 1.00000000 0.17125 0.05272110
## 10 1.77827941 0.18125 0.04093101
## 11 3.16227766 0.18750 0.04639804
## 12 5.62341325 0.18625 0.04581439
## 13 10.00000000 0.19250 0.04377975
The optimal cost is 0.3162278 and the best performance is 0.17.
radial.svm <- svm(Purchase ~ ., data = train.oj, kernel = "radial", cost = tune.out$best.parameters$cost)
train.pred <- predict(radial.svm, train.oj)
table(train.oj$Purchase, train.pred)
## train.pred
## CH MM
## CH 436 49
## MM 78 237
(49+78)/800
## [1] 0.15875
test.pred <- predict(radial.svm, test.oj)
table(test.oj$Purchase, test.pred)
## test.pred
## CH MM
## CH 150 18
## MM 32 70
(18+32)/270
## [1] 0.1851852
From tuning the svm, the training error rate increase to 0.159 and the test error rate remained the same at 0.185.
set.seed(12)
poly.svm <- svm(Purchase ~ ., data = train.oj, kernel = "poly", degree = 2)
summary(poly.svm)
##
## Call:
## svm(formula = Purchase ~ ., data = train.oj, kernel = "poly", degree = 2)
##
##
## Parameters:
## SVM-Type: C-classification
## SVM-Kernel: polynomial
## cost: 1
## degree: 2
## coef.0: 0
##
## Number of Support Vectors: 447
##
## ( 225 222 )
##
##
## Number of Classes: 2
##
## Levels:
## CH MM
447 support vectors, where 222 are level CH and 225 are level MM.
train.pred <- predict(poly.svm, train.oj)
table(train.oj$Purchase, train.pred)
## train.pred
## CH MM
## CH 449 36
## MM 110 205
(36+110)/800
## [1] 0.1825
test.pred <- predict(poly.svm, test.oj)
table(test.oj$Purchase, test.pred)
## test.pred
## CH MM
## CH 153 15
## MM 45 57
(15+45)/270
## [1] 0.2222222
The training error is 0.183 and the test error is 0.222.
set.seed(12)
tune.out = tune(svm, Purchase ~ ., data = train.oj, kernel = "poly", ranges = list(cost = 10^seq(-2,
1, by = 0.25)))
summary(tune.out)
##
## Parameter tuning of 'svm':
##
## - sampling method: 10-fold cross validation
##
## - best parameters:
## cost
## 0.5623413
##
## - best performance: 0.18375
##
## - Detailed performance results:
## cost error dispersion
## 1 0.01000000 0.37125 0.06589396
## 2 0.01778279 0.37000 0.06645801
## 3 0.03162278 0.34625 0.06822400
## 4 0.05623413 0.33375 0.06771314
## 5 0.10000000 0.28875 0.06079622
## 6 0.17782794 0.23125 0.04379958
## 7 0.31622777 0.20250 0.03944053
## 8 0.56234133 0.18375 0.04251225
## 9 1.00000000 0.19375 0.03691676
## 10 1.77827941 0.19500 0.03641962
## 11 3.16227766 0.18875 0.04185375
## 12 5.62341325 0.19250 0.02958040
## 13 10.00000000 0.20000 0.02886751
The optimal cost is 0.5623413 and the best performance is 0.184.
poly.svm <- svm(Purchase ~ ., data = train.oj, kernel = "poly", degree = 2, cost = tune.out$best.parameters$cost)
train.pred <- predict(poly.svm, train.oj)
table(train.oj$Purchase, train.pred)
## train.pred
## CH MM
## CH 447 38
## MM 111 204
(38+111)/800
## [1] 0.18625
test.pred <- predict(poly.svm, test.oj)
table(test.oj$Purchase, test.pred)
## test.pred
## CH MM
## CH 151 17
## MM 45 57
(17+45)/270
## [1] 0.2296296
The tuning of the svm resulted in a training error rate that had a very slight increase to 0.186 and an increase in test error rate to 0.2296.
The linear SVM approach on the OJ data seemed to have given the best results since it produce lower training error rates and test error rates, 0.164 and 0.149.