Original repo

Files:

• trimap.h - the entire implementation
• setup.sh - clones AnnoyLib. Run this first
• prepare_mnist_data.py and mnist_data - Mnist data from SKLearn after running through PCA. This implementation does little data preprocessing, but the original article strongly suggests running PCA if the number of dimensions is high (>100).
• demo.py - usage example of the original implementation
• test_mnist.cpp - usage example
• plots.py - compares demo.py and test_mnist results

The result of running test_mnist and plots.py: