`crossvalidation` and random search for calibrating support vector machines

[This article was first published on T. Moudiki's Webpage - R, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Install and load packages

  repos = c(techtonique = 'https://techtonique.r-universe.dev',
            CRAN = 'https://cloud.r-project.org')



Input data

transforming model response into a factor

y <- as.factor(as.numeric(iris$Species))

explanatory variables

X <- as.matrix(iris[, c("Sepal.Length", "Sepal.Width",
                     "Petal.Length", "Petal.Width")])

Objective – cross-validation – function to be maximized

OF <- function(xx) {
  res <- crossvalidation::crossval_ml(
    x = X,
    y = y,
    k = 5,
    repeats = 3,
    p = 0.8,
    fit_func = e1071::svm,
    predict_func = predict,
    packages = "e1071",
    fit_params = list(gamma = xx[1],
                      cost = xx[2])
  # default metric is accuracy

There are many, many ways to maximize this objective function.

A naive random search optimization procedure

simulation of SVM’s hyperparameters’ matrix

n_points <- 250
(hyperparams <- cbind.data.frame(
  gamma = runif(n = n_points,
                min = 0,
                max = 5),
  cost = 10 ^ runif(n = n_points,
                    min = -1,
                    max = 2)

accuracies on the set of simulated hyperparameters

scores <- parallel::mclapply(1:n_points,
                             mc.cores = parallel::detectCores())
scores <- unlist(scores)

‘best’ hyperparameters and associated training set score

max_index <- which.max(scores)
xx_best <- hyperparams[max_index,]

      gamma     cost
18 0.2102977 1.101473


|===================================================================================| 100%
utilisateur     système      écoulé 
      0.284       0.079       0.365 

[1] 0.9527778

To leave a comment for the author, please follow the link and comment on their blog: T. Moudiki's Webpage - R.

R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)