Official Code for "Continuous-time Autoencoders for Regular and Irregular Time Series Imputation" Paper in WSDM'24

install conda environments

conda create -n cta python=3.9.7
conda activate cta
sh create_env.sh

cd generating_dataset
sh data_downloading.sh
cd ..
sh dataset_generating.sh

train CTA for STOCKS 70%

VAE-AE

python main.py --config ./configs/Stocks_best/Stocks_07masked_VAE_AE.py --config.model.saving_path ./experiments/Stocks_07masked_VAE_AE_reproduce --mode train --log_path ./experiments/Stocks_07masked_VAE_AE_reproduce/train.log

AE-AE

python main.py --config ./configs/Stocks_best/Stocks_07masked_AE_AE.py --config.model.saving_path ./experiments/Stocks_07masked_AE_AE_reproduce  --mode train --log_path ./experiments/Stocks_07masked_AE_AE_reproduce/train.log

For other datasets, change dataset name and 'missing rate`

test pretrained model for STOCKS 70%

VAE-AE

python main.py --config ./configs/Stocks_best/Stocks_07masked_VAE_AE.py --config.model.saving_path ./experiments/Stocks_07masked_VAE_AE_pretrained --mode test --log_path ./experiments/Stocks_07masked_VAE_AE_pretrained/test.log

AE-AE

python main.py --config ./configs/Stocks_best/Stocks_07masked_AE_AE.py  --config.model.saving_path ./experiments/Stocks_07masked_AE_AE_pretrained --mode test --log_path ./experiments/Stocks_07masked_AE_AE_pretrained/test.log