Skip to main content
Home
Browse packagesDocsSign in
This release is 136 versions behind 0.168.1 — the latest version of @stsoftware/neat-ai. Jump to latest

@stsoftware/neat-ai@0.121.0
Built and signed on GitHub Actions

stSoftwareAU/NEAT-AI

NEAT Neural Network. This project is a unique implementation of a neural network based on the NEAT (NeuroEvolution of Augmenting Topologies) algorithm, written in DenoJS using TypeScript.

This package works with Deno
This package works with Deno
JSR Score
94%
Published
5 months ago (0.121.0)
Package root>test>Learn.ts
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778
import type { DataRecordInterface } from "../src/architecture/DataSet.ts";
import type { NeatOptions } from "../src/config/NeatOptions.ts";
import { Creature } from "../src/Creature.ts";
import { train } from "./TrainTestOnlyUtil.ts";

((globalThis as unknown) as { DEBUG: boolean }).DEBUG = true;

Deno.test("Learn", () => {
  const nn = Creature.fromJSON(
    {
      neurons: [
        {
          bias: -0.05601433047338172,
          index: 2,
          type: "hidden",
          squash: "LOGISTIC",
        },
        {
          bias: -0.03918215964297005,
          index: 3,
          type: "hidden",
          squash: "BIPOLAR",
        },
        {
          bias: 0.5402230858136775,
          index: 4,
          type: "output",
          squash: "IDENTITY",
        },
        {
          bias: -1.2019708378892324,
          index: 5,
          type: "output",
          squash: "IDENTITY",
        },
      ],
      synapses: [
        { weight: 2.0458515029017104, from: 0, to: 2 },
        { weight: -0.07677399122336755, from: 1, to: 3 },
        { weight: -0.5014045264238365, from: 1, to: 4 },
        { weight: 0.17748749525130925, from: 1, to: 5 },
        { weight: 0.0359712181205063, from: 2, to: 3 },
        { weight: -1.0963423331951507, from: 2, to: 4 },
        { weight: 2.2532403719566836, from: 2, to: 5 },
        { weight: -0.4016561292244124, from: 3, to: 5 },
      ],
      input: 2,
      output: 2,
    },
  );

  nn.fix();
  nn.validate();

  const options: NeatOptions = {
    iterations: 10000,
    log: 50,
    elitism: 3,
  };

  const dataSet: DataRecordInterface[] = [];
  for (let i = 0; i < 10; i++) {
    const input = [Math.random() * 2 - 1, Math.random() * 2 - 1];
    const dr = {
      input: input,
      output: [(input[0] + input[1]) / -2, input[0] + input[1]],
    };

    dataSet.push(dr);
  }

  const answersA = nn.activate(new Float32Array([0.1, 0.2]));
  console.info(answersA);
  train(nn, dataSet, options);

  const answersB = nn.activate(new Float32Array([0.1, 0.2]));
  console.info(answersB);
});