NanoBrain-unitypackage/Runtime/Scripts/Core/Population.cs

using System.Collections.Generic;
using UnityEngine;

namespace NanoBrain {
    public class Population {

        # region Members

        public class Member {
            public Member(Cluster cluster) {
                this.brain = cluster;
                this.performance = 0;
                this.initialized = true;
            }
            public Cluster brain;
            public float performance;
            public bool initialized;
        }
        public List<Member> members = new();

        public Member AddMember(Cluster brain) {
            Member newMember = new(brain);
            this.members.Add(newMember);
            return newMember;
        }

        #endregion Members

        #region Saving

        public enum SaveMethod {
            Average,
            Mean,
            First,
            All
        }
        public void Save(string path, SaveMethod saveMethod = SaveMethod.Average) {
            Cluster cluster = null;
            switch (saveMethod) {
                case SaveMethod.First:
                    Member member = this.members[0];
                    if (member == null)
                        return;

                    cluster = member.brain;
                    break;
                case SaveMethod.Average:
                    cluster = CalculateAverageCluster();
                    break;
            }
            cluster?.Export(path);
        }

        protected Cluster CalculateAverageCluster() {
            if (members.Count <= 0)
                return null;

            Cluster result = members[0].brain.Clone();
            for (int memberIx = 1; memberIx < members.Count; memberIx++) {
                Member member = members[memberIx];
                result.ProcessWeightsFrom(member.brain, Average);
            }

            return result;
        }

        #endregion Saving

        #region New Generation

        public void NewGeneration() {
            foreach (Member member in this.members) {
                member.initialized = false;
            }
        }

        protected List<Member> SelectElite(float percentage) {
            return SelectElite((int)(members.Count * percentage));
        }
        protected List<Member> SelectElite(int count) {
            List<Member> selectedMembers = new();

            SortMembers();
            for (int i = 0; i < count; i++) {
                Member member = this.members[i];
                if (!member.initialized) {
                    Debug.Log($"Selected {i}: {member.performance}");
                    member.initialized = true;
                    selectedMembers.Add(member);
                    // LogTrainableWeights(member.brain);
                }
            }

            return selectedMembers;
        }

        private void SortMembers() {
            members.Sort((a, b) => a.performance.CompareTo(b.performance));
        }

        protected List<Member> GenerateMutants(List<Member> elite, float percentage) {
            return GenerateMutants(elite, (int)(members.Count * percentage));
        }
        protected List<Member> GenerateMutants(List<Member> elite, int count) {
            List<Member> selectedMembers = new();
            int i = 0;
            int n = 0;
            while (n < count && i < this.members.Count) {
                Member member = this.members[i];
                if (member.initialized == false) {
                    GenerateMutant(member, elite);
                    member.initialized = true;
                    n++;
                    selectedMembers.Add(member);
                }
                i++;
            }
            return selectedMembers;
        }

        protected void GenerateMutant(Member member, List<Member> ants) {
            System.Random randomGenerator = new();
            int n = ants.Count;
            int parent1 = randomGenerator.Next(0, n);
            int parent2 = randomGenerator.Next(0, n);
            Debug.Log($"Mutate from {members[parent1].performance} and {members[parent2].performance}");

            member.brain.CopyWeightsFrom(ants[parent1].brain);
            member.brain.ProcessWeightsFrom(ants[parent2].brain, Average);
            // LogTrainableWeights(member.brain);
            member.brain.GaussianAdditiveMutation(1e-2f);
            // LogTrainableWeights(member.brain);
        }

        private static float Average(float a, float b) {
            return (a + b) / 2;
        }

        protected List<Member> GenerateRandom() {
            return GenerateRandom(int.MaxValue);
        }

        protected List<Member> GenerateRandom(int count) {
            List<Member> selectedMembers = new();
            int i = 0;
            int n = 0;
            while (n < count && i < this.members.Count) {
                Member member = this.members[i];
                if (member.initialized == false) {
                    Debug.Log($"Randomized {i}: {member.performance}");
                    member.brain.InitializeRandom();
                    member.initialized = true;
                    selectedMembers.Add(member);
                    n++;
                }
                i++;
            }
            return selectedMembers;
        }

        #endregion New Generation

        private void LogTrainableWeights(Cluster cluster) {
            string s = "";
            foreach (Nucleus nucleus in cluster.nuclei) {
                if (nucleus is Neuron neuron) {
                    foreach (Synapse synapse in neuron.synapses) {
                        if (synapse.trainable) {
                            s += synapse.weight + " ";
                        }
                    }
                }
            }
            Debug.Log(s);
        }

        public virtual void Update() { }
    }
}