A small 2 layer AI that can learn to recognize digits with over 90% accuracy.
All the algorithm does at the moment is one pass over 80% percent of the samples (~20% for testing).
Speed-wise, it can analyze well over 60,000 28 by 28 image samples in less than half a second, whether
that is good or bad is not my concern as the primary goal of this project is to learn.
Please do point out my mistakes if you have the time and let me know of any suggestions!
Once again, this is a very simple program and it does not use anything close to advanced techniques.
To use this program you should have the sample data under the 'data' directory in this source tree,
you can download the assets from: http://yann.lecun.com/exdb/mnist
At the moment I have quite a silly dependency on one of my other projects, so remember to use
--recursive when cloning.