|
| 1 | +############################################# |
| 2 | +# (c) 2014 Francesco Gadaleta # |
| 3 | +############################################# |
| 4 | + |
| 5 | +library(kernlab) |
| 6 | + |
| 7 | +data(spam) |
| 8 | +dt <- as.matrix(spam[c(10:20, 3000:3010), -58]) |
| 9 | + |
| 10 | +rbf <- rbfdot(sigma = 0.5) |
| 11 | + |
| 12 | +#rbf(o1,o2) |
| 13 | +#exp(0.5 * (2 * crossprod(o1, o2) - crossprod(o1) - crossprod(o2))) |
| 14 | +#exp(0.5 * (2 * t(o1)%*%o2 - t(o1)%*%o1 - t(o2)%*%o2)) |
| 15 | + |
| 16 | +K = kernelMatrix(kernel = rbf, x = dt) |
| 17 | +yt <- as.matrix(as.integer(spam[c(10:20, 3000:3010), 58])) |
| 18 | +yt[yt==2] = -1 |
| 19 | + |
| 20 | +# compute kernel expansion |
| 21 | +kernelMult(rbf, dt, , yt) |
| 22 | + |
| 23 | + |
| 24 | + |
| 25 | +# 1. example of ksvm |
| 26 | +data(promotergene) |
| 27 | +tindex <- sample(1:dim(promotergene)[1], 5) |
| 28 | +genetrain <- promotergene[-tindex, ] |
| 29 | +genetest <- promotergene[tindex, ] |
| 30 | + |
| 31 | +gene <- ksvm(Class ~ ., data = genetrain, kernel = "rbfdot", |
| 32 | + kpar = "automatic", C = 60, cross = 3, prob.model = TRUE) |
| 33 | +predict(gene, genetest, type="probabilities") |
| 34 | +tmp <- genetest[1, -1] |
| 35 | +#tmp["V2"] = 'a' |
| 36 | +predict(gene, tmp) |
| 37 | + |
| 38 | +# 2. example of ksvm |
| 39 | +x <- rbind(matrix(rnorm(120), , 2), matrix(rnorm(120, mean = 3), , 2)) |
| 40 | +y <- matrix(c(rep(1, 60), rep(-1, 60))) |
| 41 | +vp <- ksvm(x, y, type = "C-svc") |
| 42 | +plot(vp, data=x) |
| 43 | +newdata = matrix(c(2,0.1, 3,3), ncol = 2) |
| 44 | +predict(vp, newdata) |
| 45 | + |
| 46 | +# relevance vector machine |
| 47 | +rvmm <- rvm(x,y, kernel="rbfdot", kpar=list(sigma=0.1)) |
| 48 | +yhat <- predict(rvmm, x) |
| 49 | +plot(yhat) |
| 50 | + |
| 51 | + |
| 52 | +# ranking algorithm |
| 53 | +data(spirals) |
| 54 | +ran <- spirals[rowSums(abs(spirals) < 0.55) == 2, ] |
| 55 | +ranked <- ranking(ran, 54, kernel = "rbfdot", |
| 56 | + kpar = list(sigma = 100), edgegraph = TRUE) |
| 57 | +ranked[54,2] <- max(ranked[-54,2]) |
| 58 | +c <- 1:86 |
| 59 | +op <- par(mfrow = c(1, 2), pty = "s") |
| 60 | +plot(ran) |
| 61 | +plot(ran, cex = c[ranked[, 3]]/40) |
| 62 | + |
| 63 | + |
| 64 | +# online learning |
| 65 | +x <- rbind(matrix(rnorm(90), , 2), matrix(rnorm(90) + 3, , 2)) |
| 66 | +y <- matrix(c(rep(1, 45), rep(-1, 45)), , 1) |
| 67 | +on <- inlearn(2, kernel = "rbfdot", kpar = list(sigma = 0.2), type = "classification") |
| 68 | +ind <- sample(1:90, 90) |
| 69 | + |
| 70 | +for (i in ind) |
| 71 | + on <- onlearn(on, x[i, ], y[i], nu = 0.03, lambda = 0.1) |
| 72 | +sign(predict(on, x)) |
| 73 | + |
| 74 | +spc <- specc(x, centers=4) |
| 75 | +plot(spc@centers) |
| 76 | +plot(x) |
| 77 | + |
| 78 | + |
| 79 | + |
| 80 | + |
0 commit comments