Package: dfr 0.1.5

Fabio Feser

dfr: Dual Feature Reduction for SGL

Implementation of the Dual Feature Reduction (DFR) approach for the Sparse Group Lasso (SGL) and the Adaptive Sparse Group Lasso (aSGL) (Feser and Evangelou (2024) <doi:10.48550/arXiv.2405.17094>). The DFR approach is a feature reduction approach that applies strong screening to reduce the feature space before optimisation, leading to speed-up improvements for fitting SGL (Simon et al. (2013) <doi:10.1080/10618600.2012.681250>) and aSGL (Mendez-Civieta et al. (2020) <doi:10.1007/s11634-020-00413-8> and Poignard (2020) <doi:10.1007/s10463-018-0692-7>) models. DFR is implemented using the Adaptive Three Operator Splitting (ATOS) (Pedregosa and Gidel (2018) <doi:10.48550/arXiv.1804.02339>) algorithm, with linear and logistic SGL models supported, both of which can be fit using k-fold cross-validation. Dense and sparse input matrices are supported.

Authors:Fabio Feser [aut, cre]

dfr_0.1.5.tar.gz
dfr_0.1.5.tar.gz(r-4.5-noble)dfr_0.1.5.tar.gz(r-4.4-noble)
dfr_0.1.5.tgz(r-4.4-emscripten)dfr_0.1.5.tgz(r-4.3-emscripten)
dfr.pdf |dfr.html✨
dfr/json (API)

# Install 'dfr' in R:
install.packages('dfr', repos = 'https://cloud.r-project.org')

Bug tracker:https://github.com/ff1201/dfr/issues2 issues

On CRAN:

Conda:

2.40 score 1 stars 410 downloads 4 exports 80 dependencies

Last updated 27 days agofrom:222b23f935. Checks:3 OK. Indexed: no.

TargetResultLatest binary
Doc / VignettesOKMar 06 2025
R-4.5-linuxOKMar 06 2025
R-4.4-linuxOKMar 06 2025

Exports:dfr_adap_sgldfr_adap_sgl.cvdfr_sgldfr_sgl.cv

Dependencies:caretclasscliclockcodetoolscolorspacecpp11data.tablediagramdigestdplyre1071fansifarverforeachfuturefuture.applygenericsggplot2globalsgluegowergtablehardhatipredisobanditeratorsKernSmoothlabelinglatticelavalifecyclelistenvlubridatemagrittrMASSMatrixmgcvModelMetricsmunsellnlmennetnumDerivparallellypillarpkgconfigplyrpROCprodlimprogressrproxypurrrR6RColorBrewerRcppRcppArmadillorecipesreshape2RlabrlangrpartscalessgsshapeSLOPEsparsevctrsSQUAREMstringistringrsurvivaltibbletidyrtidyselecttimechangetimeDatetzdbutf8vctrsviridisLitewithr

Citation

To cite the DFR method in publications use

Feser F, Evangelou M (2024). “Dual feature reduction for the sparse-group lasso and its adaptive variant.” arXiv. doi:10.48550/arXiv.2405.17094, https://arxiv.org/abs/2405.17094.

To cite the dfr R package in publications use:

Feser F (2024). dfr. https://CRAN.R-project.org/package=dfr.

Corresponding BibTeX entries: 

  @Article{,
    title = {Dual feature reduction for the sparse-group lasso and its
      adaptive variant},
    author = {Fabio Feser and Marina Evangelou},
    journal = {arXiv},
    year = {2024},
    doi = {10.48550/arXiv.2405.17094},
    url = {https://arxiv.org/abs/2405.17094},
  }

  @Manual{,
    title = {dfr},
    author = {Fabio Feser},
    year = {2024},
    url = {https://CRAN.R-project.org/package=dfr},
  }

Readme and manuals

dfr

Implementation of the Dual Feature Reduction (DFR) approach for the Sparse Group Lasso (SGL) and the Adaptive Sparse Group Lasso (aSGL). The DFR approach is a feature reduction approach that applies strong screening to reduce the feature space before optimisation, leading to speed-up improvements for fitting SGL models. DFR is implemented using the Adaptive Three Operator Splitting (ATOS) algorithm, with linear and logistic SGL models supported, both of which can be fit using k-fold cross-validation. Dense and sparse input matrices are supported.

A detailed description of DFR can be found in Feser, F., Evangelou, M. (2024). "Dual feature reduction for the sparse-group lasso and its adaptive variant".

SGL was proposed in Simon, N., Friedman, J., Hastie, T., Tibshirani, R. (2013). "A Sparse-Group Lasso".

The adaptive SGL is described in Mendez-Civieta, A., Carmen Aguilera-Morillo, M., Lillo, R. (2020). "Adaptive sparse group LASSO in quantile regression" and Poignard, B. (2020). "Asymptotic theory of the adaptive Sparse Group Lasso".

Installation

You can install the current stable release from CRAN with

install.packages("dfr")

Your R configuration must allow for a working Rcpp. To install a develop the development version from GitHub run

library(devtools)
install_github("ff1201/dfr")

Example

The code for fitting a basic DFR-SGL model is:

library(dfr)
groups = c(rep(1:20, each=3),
           rep(21:40, each=4),
           rep(41:60, each=5),
           rep(61:80, each=6),
           rep(81:100, each=7))

data = sgs::gen_toy_data(p=500, n=400, groups = groups, seed_id=3)

model = dfr_sgl(X = data$X, y = data$y, groups = groups, alpha = 0.95)

where X is the input matrix, y the response vector, groups a vector containing indices for the groups of the predictors, and alpha determines the convex balance between the lasso and group lasso.

The impact of screening can be seen by turning off the screening rules:

no_screen = system.time(model <- dfr_sgl(X = data$X, y = data$y, groups = groups, alpha = 0.95,screen=FALSE))
screen = system.time(model_screen <- dfr_sgl(X = data$X, y = data$y, groups = groups, alpha = 0.95,screen=TRUE))
c(no_screen[3], screen[3])
For DFR-aSGL, run 
library(dfr)
groups = c(rep(1:20, each=3),
           rep(21:40, each=4),
           rep(41:60, each=5),
           rep(61:80, each=6),
           rep(81:100, each=7))

data = sgs::gen_toy_data(p=500, n=400, groups = groups, seed_id=3)

model = dfr_adap_sgl(X = data$X, y = data$y, groups = groups, alpha = 0.95, gamma_1 = 0.1, gamma_2 = 0.1)

where gamma_1 and gamma_2 determine the shape of the adaptive penalties. Again, we can see the impact of screening

no_screen = system.time(model <- dfr_adap_sgl(X = data$X, y = data$y, groups = groups, alpha = 0.95, gamma_1 = 0.1, gamma_2 = 0.1, screen=FALSE))
screen = system.time(model_screen <- dfr_adap_sgl(X = data$X, y = data$y, groups = groups, alpha = 0.95, gamma_1 = 0.1, gamma_2 = 0.1, screen=TRUE))
c(no_screen[3], screen[3])