contiformer.md

ContiFormer

This document gives an example on how to use the PhysioPro framework for ContiFormer: Continuous-Time Transformer for Irregular Time Series Modeling.

Run Irregular Time Series Classification

Here we take the Heartbeat classification task from UEA & UCR Time Series Classification Repository as an example.

1. Download the dataset

cd PhysioPro
mkdir data
wget http://www.timeseriesclassification.com/aeon-toolkit/Archives/Multivariate2018_ts.zip -P data

unzip data/Multivariate2018_ts.zip -d data/
rm data/Multivariate2018_ts.zip

2. Run irregular time series classification task with ContiFormer

# create the output directory
mkdir -p outputs/Multivariate_ts/Heartbeat
# run the train task
python -m physiopro.entry.train docs/configs/contiformer_mask_classification.yml --data.mask_ratio 0.3 --data.name Heartbeat
# tensorboard
tensorboard --logdir outputs/

The results will be saved to outputs/Multivariate_ts/Heartbeat directory.

3. Run regular time series classification task with ContiFormer

# create the output directory
mkdir -p outputs/Multivariate_ts/Heartbeat
# run the train task
python -m physiopro.entry.train docs/configs/contiformer_classification.yml --data.name Heartbeat
# or use the following command
python -m physiopro.entry.train docs/configs/contiformer_mask_classification.yml --data.mask_ratio 0 --data.name Heartbeat
# tensorboard
tensorboard --logdir outputs/

The results will be saved to outputs/Multivariate_ts/Heartbeat directory.

Run temporal point process

1. Download the dataset

Please download the dataset from Google Drive Link, and put it under data fold.

2. Run temporal point process task with ContiFormer

# create the output directory
mkdir -p outputs/Temporal_Point_Process/neonate
# run the train task
python -m physiopro.entry.train docs/configs/contiformer_tpp.yml
# tensorboard
tensorboard --logdir outputs/