Go to file
dadel 928c8a05ff add README and author/licence 2018-01-24 23:47:01 +01:00
README.md add README and author/licence 2018-01-24 23:47:01 +01:00
__init__.py add letters (load data letters froms images) and pnmimage (load pbm/pgm) modules 2018-01-15 23:25:31 +01:00
letters.py add README and author/licence 2018-01-24 23:47:01 +01:00
mlp.ipynb add quick and dirty implementation of mini-batches 2018-01-24 23:24:14 +01:00
mlp.py add README and author/licence 2018-01-24 23:47:01 +01:00
pnmimage.py add README and author/licence 2018-01-24 23:47:01 +01:00

README.md

Multi-Layer Perceptron

This is a Python 3 implementation of a multi-layer perceptron using Numpy and optionally Matplotlib for cost function visualization.

It is provided under GPLv3 licence.

It implements forward-propagation and backward-propagation (gradient descent).

It allows to learn examples and trying to minimize cost function.

It provides also Momentum optimization, RMSProp optimization, Adam optimization and Forbenius norm regularization.

Note: I developed this only to implement myself the algorithms for fun and not for any production purpose. I already coded MLPs years ago in C and in C++.

I tested it with images of characters (in PGM format) extracted from a scanned image with an other program I developed many years ago in C++ but very dirty so I don't provide it.

I provide the jupyter notebook I used for some tests.