Cpp code for Simulation

Bayesian Latent Factor on Image Regression with Nonignorable Missing Data

Reference

@article{wang2021bayesian,
  title={Bayesian latent factor on image regression with nonignorable missing data},
  author={Wang, Xiaoqing and Song, Xinyuan and Zhu, Hongtu},
  journal={Statistics in Medicine},
  volume={40},
  number={4},
  pages={920--932},
  year={2021},
  publisher={Wiley Online Library}
}

Getting Started

• Prerequisites:
  • Eigen 3.3.7
  • cmake 3.3.1
  • C++ 14
• Set eigen path in CmakeLists.txt by revising include_directories(<your-path>/eigen3 .)
• Compile and run as follows.

mkdir build
cd build
cmake ..
make
./exe

• Outputs
  • Bayesian estimation of replications are saved in out/rep_*.txt.
  • Posterior samples of the latest replication are saved in out/samples_*.txt

Documentation

• For changing simulation setting please refer to Class Constant including:
  • setting (model setup)
    • 1: Simulation 1 (with image)
    • 0: Simulation 2
  • ssl (Spike-and-Slab prior choice)
    • 1: Adjust SSL
    • 0: SSL
  • model_type (missing mechanism)
    • 1: Missing completely at random
    • 2: Missing at random
    • 3: Missing not at random
  • Sample size ($N$), number of eigenimages ($P$), number of repliations, true value, tunning parameters, hyperparameters etc.
• For data generation please refer to Class Generate.
• For MCMC initial values, please refer to Class Initial.
• All the C++ functions related to posterior sampling are contained in Class MCMC.

Details

• For Simulation 1
  1. Preparation (code is not provided)
     1. Matlab: read preprocessed ADNI MRIs, save as .mat.
     2. Matlab: apply svd function for singular value decomposition and acquire eigenscores and eigenimages, save as .txt.
     3. Python: create true image parameter, specify other true values, use the image (.mat) in step a to generate the dateset for simulatin 1, save as .txt.
  2. Cpp
     1. Copy and past the simulated dataset (.txt) into Cpp folder.
     2. Set setting to 1 in Constants.h.
     3. Specify simulation setting, such as sample size, number of MCMC iterfation, and true value in Constants.h and Constants.cpp.
     4. Compile and run.
• For Simulation 2
  • Set setting to 0 in Constants.h.
  • Specify simulation setting, such as sample size, number of MCMC iterfation, and true value in Constants.h and Constants.cpp.
  • Compile and run.

PS: Please feel free to contact the authors if you have any questions!