R Notebook

cost에 따른 SVM margin 비교하기

library(e1071)
set.seed(10111)
x = matrix(rnorm(40), 20, 2)
y = rep(c(-1, 1), c(10, 10))
x[y == 1, ] = x[y == 1, ] + 1
plot(x, col = y + 3, pch = 19)

dat = data.frame(x, y = as.factor(y))

그리드 만들기

make.grid = function(x, n = 75) {
    grange = apply(x, 2, range)
    x1 = seq(from = grange[1, 1], to = grange[2, 1], length = n)
    x2 = seq(from = grange[1, 2], to = grange[2, 2], length = n)
    expand.grid(X1 = x1, X2 = x2)
}
xgrid = make.grid(x)

svm의 index를 이용해서 그림그리기

svmfit_cost100 = svm(y ~ ., data = dat, kernel = "linear", cost = 100, scale = FALSE)
ygrid = predict(svmfit_cost100, xgrid)
beta = drop(t(svmfit_cost100$coefs) %*% x[svmfit_cost10$index, ])
beta0 = svmfit_cost100$rho
plot(xgrid, col = c("red", "blue")[as.numeric(ygrid)], pch = 20, cex = 0.2, main="SVM with cost = 100")
points(x, col = y + 3, pch = 19)

points(x[svmfit_cost10$index, ], pch = 5, cex = 2)
abline(beta0/beta[2], -beta[1]/beta[2])

abline((beta0 - 1)/beta[2], -beta[1]/beta[2], lty = 2)
abline((beta0 + 1)/beta[2], -beta[1]/beta[2], lty = 2)

svmfit_cost10 = svm(y ~ ., data = dat, kernel = "linear", cost = 10, scale = FALSE)
ygrid = predict(svmfit_cost10, xgrid)
beta = drop(t(svmfit_cost10$coefs) %*% x[svmfit_cost10$index, ])
beta0 = svmfit_cost10$rho
plot(xgrid, col = c("red", "blue")[as.numeric(ygrid)], pch = 20, cex = 0.2, main="SVM with cost = 10")
points(x, col = y + 3, pch = 19)

points(x[svmfit_cost10$index, ], pch = 5, cex = 2)
abline(beta0/beta[2], -beta[1]/beta[2])

abline((beta0 - 1)/beta[2], -beta[1]/beta[2], lty = 2)
abline((beta0 + 1)/beta[2], -beta[1]/beta[2], lty = 2)

svmfit_cost1 = svm(y ~ ., data = dat, kernel = "linear", cost = 1, scale = FALSE)
ygrid = predict(svmfit_cost1, xgrid)
beta = drop(t(svmfit_cost1$coefs) %*% x[svmfit_cost1$index, ])
beta0 = svmfit_cost1$rho
plot(xgrid, col = c("red", "blue")[as.numeric(ygrid)], pch = 20, cex = 0.2, main="SVM with cost = 1")
points(x, col = y + 3, pch = 19)

points(x[svmfit_cost10$index, ], pch = 5, cex = 2)
abline(beta0/beta[2], -beta[1]/beta[2])

abline((beta0 - 1)/beta[2], -beta[1]/beta[2], lty = 2)
abline((beta0 + 1)/beta[2], -beta[1]/beta[2], lty = 2)

svmfit_cost01 = svm(y ~ ., data = dat, kernel = "linear", cost = 0.1, scale = FALSE)
ygrid = predict(svmfit_cost01, xgrid)
beta = drop(t(svmfit_cost01$coefs) %*% x[svmfit_cost01$index, ])
beta0 = svmfit_cost01$rho
plot(xgrid, col = c("red", "blue")[as.numeric(ygrid)], pch = 20, cex = 0.2, main="SVM with cost = 0.1")
points(x, col = y + 3, pch = 19)

points(x[svmfit_cost10$index, ], pch = 5, cex = 2)
abline(beta0/beta[2], -beta[1]/beta[2])

abline((beta0 - 1)/beta[2], -beta[1]/beta[2], lty = 2)
abline((beta0 + 1)/beta[2], -beta[1]/beta[2], lty = 2)