Kneeliverse: A Universal Knee/ElbowDetection Library for Performance Curves

Estimating the knee/elbow point in performance curves is a challenging task. However, most of the time these points represent ideal compromises between cost and performance.

This library implements several well-known knee detection algorithms:

1. Discrete Curvature
2. DFDT
3. Kneedle
4. L-method
5. Menger curvature

Furthermore, the code in this library expands the ideas on these algorithms to detect multi-knee/elbow points in complex curves. We implemented a recursive method that allows each of the previously mentioned methods to detect multi-knee and elbow points. Some methods natively support multi-knee detection, such as:

1. Kneedle
2. Fusion
3. Z-method

Finally, we also implemented additional methods that help with knee detection tasks. As a preprocessing step, we develop a custom RDP algorithm that reduced a discrete set of points while keeping the reconstruction error to a minimum. As a post-processing step we implemented several algorithms:

1. 1D dimensional clustering, is used to merge close knee points
2. Several filters out non-relevant knees
3. Knee ranking algorithms that used several criteria to assess the quality of a knee point

Running unit tests

Several unit tests were written to validate some corner cases. The unit tests were written in unittest. Run the following commands to execute the unit tests.

python3 -m venv venv
source venv/bin/activate
python -m pip install --upgrade pip
pip install .
python -m unittest

Documentation

This library was documented using the google style docstring, it can be accessed here. Run the following commands to produce the documentation for this library.

python3 -m venv venv
source venv/bin/activate
pip install pdoc
pdoc --math -d google -o docs src/kneeliverse \
--logo https://raw.githubusercontent.com/mariolpantunes/knee/main/media/knee.png \
--favicon https://raw.githubusercontent.com/mariolpantunes/knee/main/media/knee.png

Instalation

To install the library locally, simple execute the following commands:

python3 -m venv venv
source venv/bin/activate
python -m pip install --upgrade pip
pip install .

You can also use the PyPI repository for easy access to the library:

kneeliverse>=1.0

Runing the demos

The demos can be execute as python modules using the following code:

python -m demos.curvature -i [trace]
python -m demos.dfdt -i [trace]
python -m demos.fusion -i [trace]
python -m demos.kneedle_classic -i [trace]
python -m demos.kneedle_rec -i [trace]
python -m demos.kneedle -i [trace]
python -m demos.lmethod -i [trace]
python -m demos.menger -i [trace]
python -m demos.zmethod -i [trace]

Most demos have the same parameters (with the exception of zmethod and kneedle_classic):

python -m demos.curvature -husage: curvature.py [-h] -i I [-a] [-r R] [-t T] [-c C] [-o] [-g] [-k {left,linear,right,hull}]

Multi Knee evaluation app

optional arguments:
  -h, --help            show this help message and exit
  -i I                  input file
  -a                    add even spaced points
  -r R                  RDP reconstruction threshold
  -t T                  clustering threshold
  -c C                  corner threshold
  -o                    store output (debug)
  -g                    display output (debug)
  -k {left,linear,right,hull}
                        Knee ranking method

Authors

• Mário Antunes

• Tyler Estro

License

This project is licensed under the MIT License - see the LICENSE file for details.

Copyright

This project is under the following COPYRIGHT.