Pytorch implementation of the paper "Detecting Anomalous Event Sequences with Temporal Point Processes", by Oleksandr Shchur, Ali Caner Turkmen, Tim Januschowski, Jan Gasthaus, and Stephan Günnemann, NeurIPS 2021.
-
Install the dependencies
conda env create -f environment.yml
-
Activate the conda environment
conda activate anomaly_tpp
-
Fix a bug in
tick
: There is a bug in the version oftick
distributed via PyPI that makes it incompatible withscikit-learn
.- Open the file with the bug
nano $(dirname $(which python))/../lib/python3.9/site-packages/tick/preprocessing/longitudinal_features_product.py
- Modify line 8 from
to
from sklearn.externals.joblib import Parallel, delayed
from joblib import Parallel, delayed
- Save and exit the file:
Ctrl + X
,Y
,Return
.
- Open the file with the bug
-
Install the package (this command must be run in the
tpp-anomaly-detection
folder)pip install -e .
-
Unzip the data
unzip data.zip
notebooks/spp_experiment.ipynb
: Standard Poisson process vs. other toy TPPs (Section 6.1 in the paper).notebooks/multivariate_experiment.ipynb
: Multivariate TPPs inspired by real-world scenarios (Section 6.2).notebooks/real_world_experiment.ipynb
: Real-world datasets (Section 6.3).
Please cite our paper if you use the code or the datasets in your own work
@article{
shchur2021detecting,
title={Detecting Anomalous Event Sequences with Temporal Point Processes},
author={Oleksandr Shchur and Ali Caner Turkmen and Tim Januschowski and Jan Gasthaus and and Stephan G\"{u}nemann},
journal={Advances in Neural Information Processing Systems (NeurIPS)},
year={2021},
}