Iterative Time Series Imputation by Maintaining Dependency Consistency

Datasets

This repository contains the code of IR-Square-Net and IR-Square-GAIN with four datasets. Air Quality, Human Activity, Traffic Speed are available, but Solar Energy dataset should be firstly unzipped.

Run

If you want to reproduce the experiment, please run "main.py" with Pycharm or with command

python3 main.py

The default dataset is Air Quality with 20% missing data. For the other cases, please add arguments such as

python3 main.py -model GAN -dataset traffic -r_miss 0.4 -cuda_id 0 -use_irm 1 -iter_time 2

The argument use_irm determines whether to use the incomplete representation mechanism, while iter_time represents the time of reconstruction, which are two main contributions of this work.

Contact

If you have any questions or suggestions for our paper or codes, please contact us. Email: [email protected].