This repo contains scripts to reproduce the results in the paper:
"EPEM: Efficient Parameter Estimation for Multiple Class Monotone Missing Data",
which is published in Information Sciences.
A video on motivation and data partition.
The notebooks are created by Google's Colaboratory.
The "Parameter estimation error" notebook produces the results in Table 2: Parameters estimation errors with different missing rates, except for MNIST, fashion MNIST. Meanwhile, the notebook Parameter estimation error shuffled produces the results on Table B.5: Parameters estimation errors with different missing rates on shuffled data in the Appendix except for MNIST, fashion MNIST.
The "application in linear discriminant analysis" notebook produces the results in Table 3: The cross-validation errors on datasets with different missing rates in LDA application, except for MNIST, fashion MNIST.
The folder "parameter estimation MNIST _ Fashion MNIST" contains the codes that produce the results in Table 2: Parameters estimation errors with different missing rates for MNIST, fashion MNIST.
The folder "LDA on MNIST_ Fashion MNISTT" contains the codes that produce the results in Table 3: The cross-validation errors on datasets with different missing rates in LDA application for MNIST, fashion MNIST.
We recommend you to cite our following paper when using these codes for further investigation:
@article{nguyen2021epem,
title={EPEM: Efficient Parameter Estimation for Multiple Class Monotone Missing Data},
author={Nguyen, Thu and Nguyen, Duy HM and Nguyen, Huy and Nguyen, Binh T and Wade, Bruce A},
journal={Information Sciences},
volume={567},
pages={1--22},
year={2021},
publisher={Elsevier}
}Further requests can directly be sent to the corresponding authors: Thu Nguyen (thu.nguyen@louisiana.edu) and Binh T. Nguyen (ngtbinh@hcmus.edu.vn) for an appropriate permission.