barisser / neural

Modular Neural Network Platform

Neural

Modular Neural Network Platform

This is an attempt to make an easy-to-use neural network simulation platform on Python.

Specifically, I want to enable evolutionary, genetic algorithms for neural network optimization.

Current Status

The neural network works surprisingly well

the evolutionary search is still somewhat buggy

TO DO

High level abstract way to connect separate units

In Multi-Unit systems, Unit-level I/O's exist, but system-level I/O's needed.

Testing on a system's basis with customizeable fitness function

Render image of Neural Network

Post best results online

DONE

SAVE/LOAD state of neuronal configuration in easily readable format

neural's Issues

Encapsulate Input/Output Parameters, Fitness

Need easy way for distant user to specify input/output parameters, dimensionality, and even feedback between output->next input.

Let's call it the 'environment'.

This also includes a generic fitness equation. I need an abstraction to package this, without it being hardcoded.

Saving Configurations

Need to create saving format within UNITS, across UNITS