Longitudinal Deep Kernel Gaussian Process Regression
- The raw data: we do not include the raw data file since some of them are only available after application. We instead leave the official website for downloading the raw data in the corresponding preprocessing file.
- The code for data preprocessing are included in the folder
preprocess
- The data after preprocessing are included in the folder
data
Code for our model is in file L-DKGPR
. To run our code, you should first format your data and save it using scipy.io.savemat(dataset)
where dataset
is the target data file, which contains the following fields:
- trainX: The covariate matrix for the training data. A 2D numpy array with size
$N_{train}\times P$ . - trainY: The training labels. A 1D numpy array with size
$N_{train}$ - testX: The covariate matrix for the test data. A 2D numpy array with size
$N_{test}\times P$ - testY: The testing labels. A 1D numpy array with size
$N_{test}$ - trainId: The individual id corresponding to each row of the training data.
- testId: The individual id corresponding to each row of the test data.
- trainOid: The observation id (timestamp) corresponding to each row of the training data.
- testOid: The observation id (timestamp) corresponding to each row of the test data.
To run our algorithm, you should first trypython L-DKGPR.py --help
to check the available parameters. Basically, you can just use the default setting or change the --file
to locate your target data file.
The code for our baseline models are located in baseline_model.
@inproceedings{liang2020longitudinal,
title={Longitudinal Deep Kernel Gaussian Process Regression},
author={Liang, Junjie and Wu, Yanting and Xu, Dongkuan and Honavar, Vasant},
booktitle={Proceedings of the 35th AAAI Conference on Artficial Intelligence},
year={2021}
}