I made a neural network without any external libraries.
It is currently programmed to classify flowers from the iris dataset. After each training round, it outputs 5 random data entries, the network's certainties, and the network's guess in the following format:
[type-0 certainty, type-1 certainty, type-2 certainty]
Guess: <the highest certainty>
Real: <the actual entry>
The network it uses has a sigmoid activation function, and has two size-5 hidden layers.
If you have any questions about the internals, just ask.
I wish I could make one this good. I don't like using external libraries. It always says "Accuracy: 33.3%". I wonder how you could make one that can process words. I have only been abele to make mine process numbers between 0 and 1.
If you want to make it do something different, just fork it, change the network setup, and put new data in data.py (internals don't matter, but the final result should be in DATA)
ALSO: The program has been slightly optimized.