A ML (em ell) package written in Python. The goal of this package is my own personal education. I intend to learn different machine learning techniques by implementing them myself from the ground up. This works in conjunction with my personal knowledge base at bitlog.com: [Knowledge Base].
This package is not optimized for anything but my own knowledge. Any patches / corrections / suggestions are welcome!
At the moment, it's only designed to be used in tests. The test suite can be exercised using either...
# Exercise linting and tests
tox
# Just tests
pytest
train_add
Has two demos The first has 1 neuron network with 2 inputs learning to add numbers. This is designed to be the simplest possible neural network.
The second shows that a naive implementation of neural networks doesn't
always find a good approximation for a function that it could replicate.
It tries (and fails) to learn y1 = .5 * x1 + x2 + 1
python setup.py install
train_add
MIT license. Go nuts.