mengxu98 / smtl

smTL CRAN Package

Multi-Task Learning for Sparsity Pattern Heterogeneity: A Discrete Optimiztion Approach (sMTL)

Repository Description

Repository for the development version of the R Package sMTL as well as code to run analyses and reproduce figures in the associated manuscript.

sMTL R Package

Installation

The sMTL $\texttt{CRAN}$ package can be downloaded with the following $\texttt{R}$ command:

install.packages("sMTL")

Alternatively, the development version of the Sparse Multi-Task Learning $\texttt{R}$ Package sMTL can be downloaded as follows:

library(devtools)
install_github("gloewing/sMTL", subdir = "Package/sMTL")

Package Usage

*** The first time sMTL is used, an initial setup step is required to link $\texttt{R}$ and $\texttt{Julia}$ (where the algorithms are coded) or else the package functions will not work. ***

For these steps and a tutorial on package functions, please refer to sMTL's Vignette.

Repository Folders

1. The 'Package' folder has code for the sMTL $\texttt{R}$ package that is current under development. This includes $\texttt{Julia}$ code to run algorithms and an $\texttt{R}$ wrapper code to tune and fit these models. This package is still under development and has not been widely tested.

2. The 'sMTL_Paper' folder has code to run analyses and make figures for the original (2022) version of the manuscript. Some files will require changing path names. Please feel free to reach out to the author for more annotation or help with reproducing any analyses.

3. The 'sMTL_paper_Fig1_demo' folder contains code that can be used as a short self-contained demo/introduction to many of the methods proposed in the paper. This code should reproduce Figures 1 and 3 in the manuscript. It includes $\texttt{R}$ code for data simulation, calls the algorithms (through $\texttt{Julia}$) to tune and fit models on the data and then generate figures.

4. The 'sMTL_Paper_2024' folder has code to run analyses and make figures for the updated manuscript (2024). Some files will require changing path names. Please feel free to reach out to the author for more annotation or help with reproducing any analyses.

5. The 'Exact Solvers' folder has the outer approximation (exact solver) code described in the 2024 version of the paper (and used to produce results in Table 3).

Dataset Links

Links to the data repositories for the neuroscience and cancer genomics applications can be found at:

1. Neuroscience application: https://osf.io/tb8fx/

This data was originally published with the paper:

Gabriel Loewinger, Prasad Patil, Kenneth T Kishida, Giovanni Parmigiani. (2022) Hierarchical resampling for bagging in multistudy prediction with applications to human neurochemical sensing. Annals of Applied Statistics. 16(4):2145-2165.

2. Breast Cancer application: https://osf.io/k6ynp/

This is a pre-processed version of data from:

Katie Planey. (2020). curatedBreastData: Curated breast cancer gene expression data with survival and treatment information R package version 2.18.0.