Quickest Change Detection for Unnormalized Statistical Models

This is an implementation of Quickest Change Detection for Unnormalized Statistical Models and Score-based Change Point Detection for Unnormalized Models

Requirements

See requirements.txt

Instructions

• Global hyperparameters are configured in config.yml
• Use make.sh to generate run script
• Use make.py to generate exp script
• Use process.py to process exp results
• Experimental setup are listed in make.py
• Hyperparameters can be found at process_control() in utils.py
• modules/cpd.py defines Change Point Detection methods

Examples

• Test CUSUM for MVN ( $d=2$ ) dataset with 500 pre data, 10000 post data, $\epsilon_{\mu} = 0.1$, no noise, ARL $=2000$

  python test_cpd.py MVN-2_500_10000_0.1-0.0_0_2000

• Test Scan B-statistic MVN EXP ( $d=2$ ) dataset with 500 pre data, 10000 post data, $\epsilon_{\log \sigma^2} = 0.5$, $\sigma_{noise} = 0.1$, ARL $=2000$

  python test_cpd.py MVN-2_500_10000_0.0-0.5_0.1_2000

• Test CALM-MMD for EXP ( $d=2$ ) dataset with 500 pre data, 10000 post data, $\epsilon_{\tau} = 1.0$, $\sigma_{noise} = 0.3$, ARL $=2000$

  python test_cpd.py EXP-2_500_10000_1.0_0.3_2000

• Test SCUSUM for RBM ( $d=50$ ) dataset with 500 pre data, 10000 post data, $\epsilon_{\log \sigma^2} = 0.05$, no noise, ARL $=2000$, $m=500$

  python test_cpd.py RBM-50_500_10000_0.05_0_2000_500

Results

• The results of Detection Score (before and after change) with MVN ( $\epsilon_{\mu} = 0.3$ ) and ARL $=2000$.

Acknowledgements

Suya Wu
Enmao Diao
Taposh Banerjee
Jie Ding
Vahid Tarokh