Salut mes amis et je souhaite vous etes chakhmakakh! Alors, this repository contains raw implementation of the following anomaly detection algorithms:
- Distance to All: Distance of the point to all points is calculated, and aggregated.
- Distance to Nearest Neighbor: Distance of the point to the nearest neighbor is calculated, and aggregated.
- Distance to K-Nearest Neighbors: Distance to K-Nearest neighbors is calculated, and aggregated. With this one, you can either use a named aggfunc (mean, sum, median) or define it yourself through a Lambda expression and it will be evaluated in runtime.
- K-Means Clustering: The usual K-Means deal. Generates k clusters of points, k is given. The output for the previous three is a CSV file, but for this one, it's a JSON file.
- Fuzzy C-Means Clusteinr: The fuzzy version of K-Means, this time ,the probability is calculated. The output is a chart, and not a dataset.
How to run:
-
Install Python 3.10 (and only 3.10!)
-
Create a virtual env with
python3.10 -m virtualenv [envname] -
source [envname]/bin/activate(On Linux)
Now you have two ways to run the application.
A) Through commandline
Just run python3 ad.py -h to see the help pop up, then decide the parameters. I don't see the need to explain
every thing here because the help explains everything clearly.
B) Edit params.json and run python3 adjson.py --- keep in mind that you can give it path to another
params.json like python3 adjson ./params2.json.
Inside params.json there's a parameter called python_executable. If you are using
virtualenv, and you have currently activated it, set it to python3. Otherwise, give it the path to your Python 3.10 installation!
If you wish to use a default value, use DEFAULT or default.
Any questions, direct it at Chubak#7400 on Discord.
You can use the examples file to test things out.
Distance functions are from SciPy but the main algorithms are by me using np.vectorize, fastest possible!
Et alors je bid a revioir!