ARTtransfer {ARTtransfer}R Documentation

ARTtransfer: Adaptive and Robust Transfer Learning for Enhanced Model Performance

Description

The ARTtransfer package implements the Adaptive and Robust Transfer Learning (ART) framework introduced by Wang et al. (2023, DOI: 10.1002/sta4.582). This framework enhances model performance on a primary task (target domain) by utilizing information from auxiliary datasets (source domains). The package is specifically designed to prevent negative transfer, ensuring that auxiliary data does not degrade model performance.

Details

ARTtransfer is a flexible and general framework that also includes variable importance metrics, enabling users to evaluate the contribution of each variable to the final model and improve the interpretability of results.

The package includes implementations of common predictive models such as linear regression, logistic regression, lasso and elastic net penalized regression (both linear and logistic), random forest, gradient boosting machines, and neural networks. Users can also define and integrate their own predictive models into the ART framework, based on the provided examples.

In addition, the package implements a framework ART-I-AM that aggregates multiple machine learning methods into the ART process. This is exemplified with implementations of random forests, AdaBoost, and a basic neural network. Users can further expand the framework by incorporating other predictive models of their choice.

Functions

- ART(): Main function for performing adaptive and robust transfer learning. - generate_data(): Generates synthetic datasets for transfer learning simulations. - Wrapper functions: Functions like fit_lm(), fit_logit(), and fit_random_forest() are used to fit models in the ART framework.

Examples

To perform ART on synthetic data: dat <- generate_data(n0=100, K=3, nk=50, is_noise=TRUE, p=10) fit <- ART(dat$X, dat$y, dat$X_aux, dat$y_aux, dat$X_test, func=fit_lm, lam=1)

For more details, see the documentation for individual functions.

Author(s)

Maintainer: Boxiang Wang boxiang-wang@uiowa.edu

Authors:


[Package ARTtransfer version 1.0.0 Index]