This project is based on the Numenta Anomaly Benchmark and provides a testing environment for multivariate streaming anomaly detection models.
Three models have been implemented:
- The subspace tracking model SPIRIT, by Papadimitriou et al. link
- The kernel mean embedding model EXPoSE, by Schneider et al. link
- A custom Conditional kernel mean embedding Conditional EXPoSE, which was implemented for a master's thesis.
Next to certain original NAB datasets, additional synthetic datasets were generated.
- Put your dataset in
data-preprocessed/ - Follow
src/pipeline.py - Results should be shown and stored in
results/