Awesome

Contextual Point-Process Outlier Detection (CPPOD)

This code accompanies the paper:

Event Outlier Detection in Continuous Time. Siqi Liu, and Milos Hauskrecht. International Conference on Machine Learning, 2021.

Dependencies

The code runs on Python 3.6 or 3.7. It relies on the following libraries and has been tested on the versions shown, although newer versions may still work:

• python=3.6.10
• pytorch=1.7.0
• numpy=1.18.5
• scipy=1.5.4
• statsmodels=0.11.1
• scikit-learn=0.19.2
• pandas=0.23.4
• matplotlib=2.2.3

Reproducing results

Steps:

• Make directories
  • data/pois
  • data/gam
  • result/fig
  • result/tab
• Generate data
  • python simulate_data_pois.py (Poisson process)
  • python simulate_data_gam.py (Gamma process)
• Run baselines
  • python train_test_baselines.py
• Train and test CPPOD and PPOD
  • Run the commands in train_test_cppod_sim.sh
• Evaluate the performance
  • ROC curves: python summarize_results.py
  • AUROC tables: python summarize_results_std.py
  • Bounds: python verify_bounds.py (some figures are not used and generated only for convenience)