Description:
This is the code related to the streamRHF paper, which was a collaboration between Télécom Paris, Huawei Technologies France, and Inria Paris.
streamRHF is an unsupervised anomaly detection algorithm for data streams. Our algorithm builds on some of the ideas of Random Histogram Forest (RHF), a state- of-the-art algorithm for batch unsupervised anomaly detection. Our approach is tree-based which boasts several appealing properties, such as explainability of the results. We conduct an extensive experimental evaluation on multiple datasets from different real-world applications. Our evaluation shows that our streaming algorithm achieves comparable average precision to RHF while outperforming state-of-the-art streaming approaches for unsupervised anomaly detection with furthermore limited computational complexity.
Other things to include:
- Technology stack: Cython
- Status: The code was completed for the publication and is no longer updated.
Install all dependencies as follows:
pip install -r requirements.txt
If you attempted to previously install the module, you may use make clean to clear all build and cache files.
- Build the streamRHF module:
make - Install the module to your local user:
make install
An example script found in scripts/insertion.py is provided that details how the streamRHF module was used in the experiments detailed in the paper.
- Navigate to the scripts directory:
cd scripts/ - Create a file named
config.iniwith the following structure:
[DATA]
dataset_path=your_path_to_datasets
- Test streamRHF on a batch or time series dataset:
python3 insertion.py [dataset] [T] [H] [iterations] [initsample] [shuffled?] [constant?]
- dataset: the name of the file containing your data
- T: the number of trees
- H: the height
- iterations: the number of iterations
- initsample: the initial sample size or window size
- shuffled: set to 1 if you want to shuffle the dataset, otherwise 0
- constant: set to 1 if the initial sample is expressed as a constant, otherwise it is a percentage of the dataset size
Example on a shuffled dataset named "cardio" with 100 trees of height 5 running on 10 iterations and an initial sample size of 5 percent:
python3 insertion.py cardio 100 5 10 5 1 0
Contact the author at stefan <dot> nesic <at> protonmail <dot> com.