neated

Description:

neated

NeuroEvolution of Augmenting Topologies Expressed in Dart! Create and evolve a neural net for solving all (or at least some) of life's hard problems.
Features #

Serialize your neural net to save and restore
Doesn't require the tracking of innovation numbers, meaning you can run disconnected and still crossover later
Includes the mutation of activation functions

Getting started #

Create a Neural Network with a certain number of inputs/outputs
Generate a new generation
Iterate through each genome of the generation
Set the inputs and call update, then get the outputs
Set the fitness of each genome and set it's fitness
Go back to Step 2 and continue until the fitness is good enough

Usage #
Here's an example of teaching a network to count from 0 to 9
var nn = NeuralNet(0, 1);
nn.createNextGeneration();
num currentFitness = 0;

for (int i=0; i<100; i++) {
for (var g in nn.currentGeneration) {
num fitness = 0;
for (int i=0;i<10;i++) {
var target = i;
g.update();
var result = g.getOutputs().toList().single;
fitness += Activation.gaussian(result - target);
}
g.fitness = fitness;
}
nn.createNextGeneration();
}


copied to clipboard