Autopopulus

An implementation of autoencoder imputation from our paper "Autopopulus: A Novel Framework for Autoencoder Imputation on Large Clinical Datasets".

Summary

Usage of the AEImputer class is similar to sklearn's imputers. Both CommonDataModule and AEImputer can be partially initialized using command-line arguments. For more examples, explore task_logic/ae_experiments.py.

First initialize a CommonDataModule object for your dataset and call setup() on it. After initializing AEImputer with the settings you like, to train the imputer call aeimputer_object.fit(data_object). To impute with the imputer call aeimputer_object.transform(dataset), where the dataset is expected to be a numpy array or a pandas dataframe.

To run experiments, edit hyperparameters in guild.yml and then run python imputer.py. To tune hyperparameters edit the config ranges in tuner.py and then call run_tune() in your routine.

References

All methods are implemented natively, but for reference:

• Pytorch implemention of MIDA available by Gondara L., et. al..
• Implementation of DAPS by Beaulieu-Jones et. al..
• Implementation of IFAC-VAE by McCoy et. al..

Collaborators

Parts of code either additionally reviewed or provided by Panayiotis Petousis and Tyler Davis.

Citation

If you use this framework in your work please cite it

@INPROCEEDINGS{9630135,
  author={Zamanzadeh, Davina J. and Petousis, Panayiotis and Davis, Tyler A. and Nicholas, Susanne B. and Norris, Keith C. and Tuttle, Katherine R. and Bui, Alex A. T. and Sarrafzadeh, Majid},
  booktitle={2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC)}, 
  title={Autopopulus: A Novel Framework for Autoencoder Imputation on Large Clinical Datasets}, 
  year={2021},
  volume={},
  number={},
  pages={2303-2309},
  doi={10.1109/EMBC46164.2021.9630135}}