Classifying the MNIST handwritten digits with a random forest in C++. Multithreaded!
- 500 decision trees
- 1 000 random training images per tree
- 28 random pixels per tree (root of the number of pixels per image)
This results in an average accuracy of 91%.
- Training 60 000 images: 40s
- Classifying 10 000 images: 340s
(tested with an i7-4820K)
Parser from https://github.com/wichtounet/mnist