/MIDA-pytorch

PyTorch implementation of "MIDA: Multiple Imputation using Denoising Autoencoders"

Primary LanguageJupyter NotebookMIT LicenseMIT

MIDA-pytorch

A pytorch implementation of "MIDA: Multiple Imputation using Denoising Autoencoders"

Summary

  1. Doing imputation with Overcomplete AutoEncoder for missing data
  2. Using complete data for training
  3. Dropout is used to generate artificial missings in the training session
  4. Experimenting with two missing methods(MCAR/MNAR)
  5. Simple but good

Requirements

  • python==3.6
  • numpy==1.14.2
  • pandas==0.22.0
  • scikit-learn==0.19.1
  • pytorch==1.0.0

Data

In the paper, 15 publicly available datasets used.
In this code, only 'Boston Housing' data is used among 15.
http://math.furman.edu/~dcs/courses/math47/R/library/mlbench/html/BostonHousing.html