daneel

Machine Learning scripts to train agents in unity fast.

How to use

First of all, you have to configure the constants in MindConstants with the number of input nodes, output nodes and hidden nodes. You may set your preferred mutation rate and other desirable settings.

Make your playable GameObject extend from NetInterface and implement the "Think" function where you should program your model's inputs and outputs.

You must have something like that:

public class PlayableObject : NetInterface
{
    // Object controller
  
    protected override IEnumerator Think()
    {
        float[] inputs;
        while (game_running)
        {
            inputs = new float[] {
                your,
                inputs
            };
            outputs = mind.Compute(inputs);
            
            // Do something with your outputs
           
            yield return new WaitForSeconds(think_frequency);
        }
    }
}

Then, drag Genetic to the scene and write your learning algorithm in your GameController's script. For example:

private Genetic genetic;
private GameObject[] samples;
    
private void Start()
{
    genetic.GenerateRandomGeneration();
    samples = genetic.Samples;
    StartCoroutine(Learn());
}    

private IEnumerator Learn()
{
    while (true)
    {
        for (int i = 0; i < samples.Length; i++)
        {
            samples[i].SetActive(true);
            game_running = true;
            while (game_running)
            {
                yield return new WaitForEndOfFrame();
            }
            samples[i].SetActive(false);
            if (samples[i].GetComponent<NetInterface>().Score < new_score)
                samples[i].GetComponent<NetInterface>().Score = new_score;
            ResetGame();
            Debug.Log("Score of sample " + samples[i].GetComponent<NetInterface>().Id + ": " 
                + samples[i].GetComponent<NetInterface>().Score + "\nNext sample incoming...");
        }
        genetic.NextGeneration();
        genetic.SaveBestSample("sample_" + genetic.Best_sample.GetComponent<NetInterface>().Id + ".txt");
        Debug.Log("Best sample: " + genetic.Best_sample.GetComponent<NetInterface>().Id 
            + "\nNext generation in coming.");
        samples = genetic.Samples;
    }
}

Then, drag your prefab model to the Genetic script's Model field in the Unity's inspector and Play.

Save your samples

With the function 'SaveBestSample'

genetic.SaveBestSample("sample_" + genetic.Best_sample.GetComponent<NetInterface>().Id + ".txt");

you can save the best sample in a text file that you can then load to continue with the training if it was stopped.

To load a sample, instead of using

genetic.GenerateRandomGeneration();

to generate a random generation, use

genetic.GenerateGenerationFromModel("your_samples_name.txt");