T-Rex

T-Rex is an evolutionary neural network. It learns by adjusting the strength of the connection weights by mutation and selection. The programmer must define the problem to solve with a scoring system so that T-Rex can evolve gradually until finding the optimal solution.

Main features:

Binary feedforward neural network
Configurable number of inputs, hidden layers and outputs
Developed using object-oriented programming
Fast, robust and portable

Network topology

Feedforward neural network: each neuron in one layer has directed connections to the neurons of the subsequent layer
Binary neural network: the input and output values can be 0 or 1 and the connection weights can be -1 or 1
Configurable number of inputs
Configurable number of hidden layers
The number of neurons in each hidden layer is set to be the number of input neurons
Configurable number of outputs

Installing dependencies

T-Rex uses the GNOME JSON-Glib library to load trained neural networks from json files:

https://wiki.gnome.org/Projects/JsonGlib

To install JSON-Glib in Debian, open a system console and run this command:

$ sudo apt install libjson-glib-dev

Building the included examples

Edit the file RunExample.c and set the name of the selected example in the SELECTED_EXAMPLE macro.

Run this command to build the binary file:

$ make

Run this command to start the selected example:

$ ./runExample

Building a shared library

Run this command to build a shared library in the current folder:

$ make library=true

T-Rex is compiled with -fshort-enums by default. If there are negative values the enum type is the first of char, short and int that can represent all the values, otherwise it is the first of unsigned char, unsigned short and unsigned int that can represent all the values.

Cleaning

Run this command to delete the generated files:

$ make clean