s2net-package {s2net}

The Generalized Semi-Supervised Elastic-Net

Description

Implements the generalized semi-supervised elastic-net. This method extends the supervised elastic-net problem, and thus it is a practical solution to the problem of feature selection in semi-supervised contexts. Its mathematical formulation is presented from a general perspective, covering a wide range of models. We focus on linear and logistic responses, but the implementation could be easily extended to other losses in generalized linear models. We develop a flexible and fast implementation, written in 'C++' using 'RcppArmadillo' and integrated into R via 'Rcpp' modules. See Culp, M. 2013 <doi:10.1080/10618600.2012.657139> for references on the Joint Trained Elastic-Net.

Details

The DESCRIPTION file:

Package:	s2net
Type:	Package
Title:	The Generalized Semi-Supervised Elastic-Net
Version:	1.0.7
Date:	2024-03-04
Authors@R:	c(person("Juan C.", "Laria",, role = c("aut", "cre"), email = "[e-mail]", comment = c(ORCID = "0000-0001-7734-9647")), person("Line H.", "Clemmensen",, role = c("aut"), email = "[e-mail]"))
Description:	Implements the generalized semi-supervised elastic-net. This method extends the supervised elastic-net problem, and thus it is a practical solution to the problem of feature selection in semi-supervised contexts. Its mathematical formulation is presented from a general perspective, covering a wide range of models. We focus on linear and logistic responses, but the implementation could be easily extended to other losses in generalized linear models. We develop a flexible and fast implementation, written in 'C++' using 'RcppArmadillo' and integrated into R via 'Rcpp' modules. See Culp, M. 2013 <doi:10.1080/10618600.2012.657139> for references on the Joint Trained Elastic-Net.
License:	GPL (>= 2)
Imports:	Rcpp, methods, MASS
Depends:	stats
LinkingTo:	Rcpp, RcppArmadillo
Suggests:	knitr, rmarkdown, glmnet, Metrics, testthat
VignetteBuilder:	knitr
URL:	https://github.com/jlaria/s2net
BugReports:	https://github.com/jlaria/s2net/issues
Encoding:	UTF-8
RoxygenNote:	7.2.0
Author:	Juan C. Laria [aut, cre] (<https://orcid.org/0000-0001-7734-9647>), Line H. Clemmensen [aut]
Maintainer:	Juan C. Laria <[e-mail]>

Index of help topics:

Rcpp_s2net-class        Class 's2net'
auto_mpg                Auto MPG Data Set
predict.s2netR          S3 Methods for 's2netR' objects.
predict_Rcpp_s2net      Predict method for 's2net' C++ class.
print.s2Data            Print methods for S3 objects
s2Data                  Data wrapper for 's2net'.
s2Fista                 Hyper-parameter wrapper for FISTA.
s2Params                Hyper-parameter wrapper for 's2net'
s2net                   The Generalized Semi-Supervised Elastic-Net
s2netR                  Trains a generalized extended linear joint
                        trained model using semi-supervised data.
simulate_extra          Simulate extrapolated data
simulate_groups         Simulate data (two groups design)

This package includes a very easy-to-use interface for handling data, with the s2Data function. The main function of the package is the s2netR function, which is a wrapper for the Rcpp_s2net (s2net) class.

Author(s)

Juan C. Laria [aut, cre] (<https://orcid.org/0000-0001-7734-9647>), Line H. Clemmensen [aut]

References

Laria, J.C., L. Clemmensen (2019). A generalized elastic-net for semi-supervised learning of sparse features.

Sogaard Larsen, J. et. al. (2019). Semi-supervised covariate shift modelling of spectroscopic data.

Ryan, K. J., & Culp, M. V. (2015). On semi-supervised linear regression in covariate shift problems. The Journal of Machine Learning Research, 16(1), 3183-3217.

See Also

s2Data, s2netR, Rcpp_s2net

Examples

data("auto_mpg")
train = s2Data(xL = auto_mpg$P1$xL, yL = auto_mpg$P1$yL,  xU = auto_mpg$P1$xU)

model = s2netR(train, 
                s2Params(lambda1 = 0.1, 
                           lambda2 = 0,
                           gamma1 = 0.1,
                           gamma2 = 100,
                           gamma3 = 0.1))

# here we tell it to transform the valid data as we did with train.
valid = s2Data(auto_mpg$P1$xU, auto_mpg$P1$yU, preprocess = train) 
ypred = predict(model, valid$xL)

## Not run: 
if(require(ggplot2)){
  ggplot() + 
    aes(x = ypred, y = valid$yL) + geom_point() + 
    geom_abline(intercept = 0, slope = 1, linetype = 2)
}

## End(Not run)