{maize} package

maize

{maize} 🌽 is an extension library for kernels & support vector machines in tidymodels! The package consists of additional kernel bindings that are not available in the {parsnip} or {recipes} package. Many of the kernels are ported from {kernlab}, additional kernels have been added directly to maize transposed from Python and Julia packages.

{parnsip} has three kernels available: linear, radial basis function, & polynomial. {maize} extends to further kernels, other engines, and adds steps for {recipes}:

Installation

You can install the development version of maize from GitHub with:

# install.packages("pak")
pak::pak("frankiethull/maize")

engines

{kernlab}

SVMs with Specialty Kernels. Contains additional regression and classification techniques such as LS-SVMs.

{kernlab} bindings🌽
extension	maize	engine	mode
{parsnip}	svm_laplace	kernlab::ksvm	regression & classification
{parsnip}	svm_tanh	kernlab::ksvm	regression & classification
{parsnip}	svm_bessel	kernlab::ksvm	regression & classification
{parsnip}	svm_anova_rbf	kernlab::ksvm	regression & classification
{parsnip}	svm_spline	kernlab::ksvm	regression & classification
{parsnip}	svm_cossim	kernlab::ksvm	regression & classification
{parsnip}	svm_cauchy	kernlab::ksvm	regression & classification
{parsnip}	svm_tanimoto	kernlab::ksvm	regression & classification
{parsnip}	svm_sorenson	kernlab::ksvm	regression & classification
{parsnip}	svm_tstudent	kernlab::ksvm	regression & classification
{parsnip}	svm_fourier	kernlab::ksvm	regression & classification
{parsnip}	svm_wavelet	kernlab::ksvm	regression & classification
{parsnip}	svm_string	kernlab::ksvm	classification
{parsnip}	lssvm_laplace	kernlab::lssvm	classification
{parsnip}	rvm_laplace	kernlab::rvm	regression
{parsnip}	kqr_laplace	kernlab::kqr	regression

{mildsvm}

Multi-Instance Learners with SVMs. In particular, MIL with ordinal outcomes using One-vs-All.

{mildsvm} bindings🌽
extension	maize	engine	mode
{parsnip}	misvm_orova_rbf	mildsvm::misvm_orova	classification

{ebmc}

Bagging and Boosting weak learners via Random Under Sampling for binary classification.

{ebmc} bindings🌽
extension	maize	engine	mode
{parsnip}	bag_svm_laplace	ebmc::ub	binary-classification
{parsnip}	rus_boost_svm_laplace	ebmc::rus	binary-classification
{parsnip}	ada_boost_svm_laplace	ebmc::adam2	binary-classification

recipes

Steps for feature engineering data via kernel related methods.

pre-processors🌽
extension	maize	engine	mode
{recipes}	step_kpca_laplace	kernlab::kpca	transformation steps
{recipes}	step_kpca_tanh	kernlab::kpca	transformation steps
{recipes}	step_kha_laplace	kernlab::kha	transformation steps
{recipes}	step_kha_tanh	kernlab::kha	transformation steps
{recipes}	step_kfa_laplace	kernlab::kfa	transformation steps
{recipes}	step_kfm_nystrom	mildsvm::kfm_nystrom	transformation steps

probably

Point calibration and conformal quantile regression with SVMs (QRSVM) for prediction intervals.

post-processors🌽
extension	maize	engine	mode
{probably}	int_conformal_quantile_svm	qrsvm::qrsvm	prediction intervals
{probably}	cal_estimate_svm	kernlab::ksvm	calibrator

modeltime

A special implementation of SVMs for time series regression. ARIMA & AutoARIMA with SVM Errors are registered in maize. The harvestime vignette showcases how-to-use ARIMA-SVMs & Recursive SVMs.

{modeltime} bindings🌽
extension	maize	engine	mode
{modeltime}	arima_svm_laplace	forecast::Arima + maize::svm_laplace	regression

applicable

One-Class SVMs for novelty detection

applicability🌽
extension	maize	engine	mode
{applicable}	apd_svm_novel_detection	kernlab::ksvm	one-SVC novelty detection

corrr

Inspired by corrr, returns a tidy data frame class (kcor_df) for Kernel Canonical Correlation Analysis:

KCCA🌽
extension	maize	engine	mode
{corrr}	kcca_correlate	kernlab::kcca	kernel canonical correlation analysis