This is a python implimentation of neural networks with back propogation and experimental stability metrics to speed convergence of cortex training
This was created as my own learning excersize for neural networks. It manages the creation and training of neural networks or "cortex's" with some diagnostic and tracking ability. It has thus far been tested to train cortex's on fairly simple training datasets, but has performed extremely reliably.
Users wishing to learn a little bit about neural networks can browse through the methods of the "neuron" class first, and create a script or two to manually pass input and output out of a single neuron with simple learning before moving up to use the "cortex" class to manage groups of neurons with more complex learning by back propogation.
hope you find it informative, enjoy!
Does not require installation, presently set up only to work in local directory
Requires python 2.7
What this package NEEDS is testing on more complex datasets, and a new method or two for loading lots of inputs into a trained cortex and saving the outputs with relevant statistical information. It is presently built to import training data via one of two methods.
- from a text file with the following format with a header (note the semicolons)
in_name1, in_name2, ... , in_nameN ; out_name1, out_name2, ... , out_nameN
in1, in2, ..., inN; out1, out2, ..., outN
in1, in2, ..., inN; out1, out2, ..., outN
in1, in2, ..., inN; out1, out2, ..., outN etc...
- from two separate csv files (inputs and targets) with an equal number of rows.
Python 2.7