RandomForest

Classifying the MNIST handwritten digits with a random forest in C++. Multithreaded!

Default parameters

• 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%.

Performance

• Training 60 000 images: 40s
• Classifying 10 000 images: 340s

(tested with an i7-4820K)

---

Parser from https://github.com/wichtounet/mnist