A feed forward perceptron network written in raw C89. It scores 86% accuracy on the MNIST database of handwritten numbers (included).
sample execution: ./ffpnet 10 0.0002
This execution will both train and test the net, as well as exporting the net for later use.
This project was previously affectionately known as Arwen
Sample output:
./ffpnet 10 0.0002
percent correct So Far:-nan this batch: 0.00
percent correct So Far:0.14 this batch: 0.14
percent correct So Far:0.14 this batch: 0.15
percent correct So Far:0.17 this batch: 0.23
percent correct So Far:0.25 this batch: 0.49
percent correct So Far:0.29 this batch: 0.46
percent correct So Far:0.32 this batch: 0.42
percent correct So Far:0.35 this batch: 0.54
percent correct So Far:0.37 this batch: 0.54
percent correct So Far:0.40 this batch: 0.61
percent correct So Far:0.43 this batch: 0.72
...
percent correct So Far:0.84 this batch: 0.98
percent correct So Far:0.84 this batch: 0.95
percent correct So Far:0.84 this batch: 0.79
percent correct So Far:0.84 this batch: 0.98
percent correct (training): 0.84
percent correct (testing): 0.86
(8583 out of 10000)