Model Averaging-Assisted Optimal Transfer Learning.

matrans

This package provides prediction tools under multi-source transfer learning framework based on frequentist model averaging strategy. It is primarily built on statistical model frameworks, including linear regression models, partially linear models. Unlike existing approaches, the proposed methods can adaptively integrate the auxiliary information from different sources and possess asymptotic optimality for prediction on the target model. It is worth noting that this package is the first open-source software package for transfer learning based on the optimal model averaging methods, providing convenient and efficient computational tools for practitioners in multi-source data modeling. For specific details, please refer to the following literature:

[1] Hu, X., & Zhang, X. (2023). Optimal Parameter-Transfer Learning by Semiparametric Model Averaging. Journal of Machine Learning Research, 24(358), 1-53.

Any questions or comments, please don’t hesitate to contact with me any time.

Installation

You can install the development version of the package like so:

library("devtools")
devtools::install_github("XnhuUcas/matrans")

Maintainer

Xiaonan Hu ([email protected])

Usage

This is a simple example which shows users how to use the provided functions to estimate weights and make predictions.

First, we generate simulation datasets (M=3) under the corrected target model and homogeneous dimension settings.

library(matrans)

## generate simulation datasets (M=3)
size <- c(150, 200, 200, 150)
coeff0 <- cbind(
  as.matrix(c(1.4, -1.2, 1, -0.8, 0.65, 0.3)),
  as.matrix(c(1.4, -1.2, 1, -0.8, 0.65, 0.3) + 0.02),
  as.matrix(c(1.4, -1.2, 1, -0.8, 0.65, 0.3) + 0.3),
  as.matrix(c(1.4, -1.2, 1, -0.8, 0.65, 0.3))
)
px <- 6
err.sigma <- 0.5
rho <- 0.5
size.test <- 500

whole.data <- simdata.gen(
  px = px, num.source = 4, size = size, coeff0 = coeff0, coeff.mis = as.matrix(c(coeff0[, 2], 1.8)),
  err.sigma = err.sigma, rho = rho, size.test = size.test, sim.set = "homo", tar.spec = "cor",
  if.heter = FALSE
)
data.train <- whole.data$data.train
data.test <- whole.data$data.test

Then, we apply the functions to implement weights estimation and out-of-sample predictions.

## optimal weights estimation
bs.para <- list(bs.df = rep(3, 3), bs.degree = rep(3, 3))
data.train$data.x[[2]] <- data.train$data.x[[2]][, -7]
fit.transsmap <- trans.smap(train.data = data.train, nfold = 5, bs.para = bs.para)
ma.weights <- fit.transsmap$weight.est
time.transsmap <- fit.transsmap$time.transsmap

## out-of-sample prediction results
pred.res <- pred.transsmap(object = fit.transsmap, newdata = data.test, bs.para = bs.para)
pred.val <- pred.res$predict.val