KnnGL

Author

Name: Deep Majumder
Roll Number: 19CS30015

Build

You need to have OpenGL libraries and headers installed. Additionally, you'd need a working installation of Python. After this, to build the Python library:

mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make

This will create knngl.cpython-<major><minor>-<arch>-<os>-<abi>.so. So on amd64 Linux with Python 3.11, this would be knngl.cpython-311-x86_64-linux-gnu.so. This shared library can be placed in any directory and we can simply use knngl by importing import knngl.

Running

In order to run knngl, we need a "fake" screen to setup the OpenGL context. This can be done using Xvfb for XOrg.

Xvfb :99 -screen 0 1920x1080x24

Thereafter, you can set the DISPLAY environment variable to :99 to set the fake screen.

DISPLAY=:99 python <file>.py

Tests

Relevant tests are copied to the build/ directory when invoking cmake. They must be run from the build/ directory itself to ensure Python finds the knngl library. There are two tests:

test.py: This is a trivial test and checks that KnnGL and Scikit return the same neighbours array.
test_adult.py: This test runs KNN on the Adult dataset using both KnnGL and Scikit and prints out the classification accuracy.

Benchmarks

Relevant benchmarks are copied to the build/ directory when invoking cmake. They must be run from the build/ directory itself to ensure Python finds the knngl library. There are two benchmarks:

bench_adult.py: Runs the benchmark on Adult dataset and prints time taken for both KnnGL and Scikit.
bench_adult_rpi.py: Same benchmarks as above but trimmed to run on the Raspberry Pi.

RedDocMD / knngl