microCPD

Micro change point detection algorithms for IoT devices.

Codes and experiments results for a research paper: "Energy-Efficient Change Point Detection Algorithm for Resource-Constrained Devices"

Results

Experiments results can be found in the results directory and other specified locations:

1. Ablation Study:
   • Detailed logs of the ablation study can be found in results/ablation.txt.
   • A summary of the ablation speedup analysis is available in algorithms/ablation_speedup_analysis.csv. The analysis is performed by the analyze_ablation_results function in algorithms/ablation_runner.py.
2. Hyperparameter Settings:
   • Hyperparameter tuning summaries for various datasets are located in the results/hyper_tuning/ directory (e.g., results/hyper_tuning/apple_summary.txt, results/hyper_tuning/measles_summary.txt).
3. Energy Consumption:
   • Energy consumption results for ESP32 devices are in the results/esp32_energy directory.
   • Energy consumption results for Raspberry Pi devices are in the results/pi_energy directory.

Algorithm Implementation

Algorithm implementations for microWATCH and other baseline Change Point Detection (CPD) algorithms (e.g., bocpd.py, cusum.py, pelt.py, micro_watch.py) are in the algorithms/ directory.

CPDPerf Implementation

Code for implementing the CPDPerf framework is in the host/ and target/ directories (for host and different kinds of devices). For example, detector implementations for target devices can be found in files