This directory contains the code to reproduce the simulation results in the manuscript "Transfer Learning with Uncertainty Quantification: Random Effect Calibration of Source to Target (RECaST)". Please contact Jimmy Hickey at jhickey@ncsu.edu for any help or questions.

See pipeline.sh for line-by-line Unix code for reproducing the results.

colmeans_missing.jl

• Contains a function to take column means of arrays with missing values.

continuous_nn.jl

• Contains a function to build a continuous neural network in Flux.

discrete_nn.jl

• Contains a function to build a discrete neural network in Flux.

expit.jl

• Contains functions to take the expit and logit of a number.

glm_regression.jl

• Contains function to fit linear of logistic regression for source models.

make_directories.jl

• Contains functions to build the output directory structure.

mse.jl

• Contains a function to calculate the MSE between a vector of true values and a vector of predicted values.

pipeline.sh

• bash file that runs the whole simulated data analysis.

posterior_predictive.jl

• Contains a function to calculate the posterior predictive prediction metrics (RMSE and AUC) as well as the continuous coverage.

prepare_simulated_data.jl

• Contains a function that generates simulated data for given sample size and noise.

recast_binary_coverage.jl

• Contains a function to calculate the coverage for a binary response.

roc.jl

• Contains a function to calculate the AUC and ROC given predicted probabilities.

run_file.jl

• Runs the entire pipeline including source models, target only models, and both RECaST models.

run_wiens_glmnet.jl

• Contains a wrapper function that runs the wiens_method_glmnet.R script.

theta_S.csv

• Saved source covariates for data generation.

train_nn.jl

• Contains a function to train neural networks using Flux.

wiens_method_glmnet.R

• Contains a function to penalized logistic regression using the glmnet package in R.