Exemplo n.º 1
0
    void Crossover()
    {
        if (bestPlayers.Count > 1)
        {
            int iParentA = Random.Range(0, bestPlayers.Count);
            int iParentB;
            do
            {
                iParentB = Random.Range(0, bestPlayers.Count);
            }while (iParentA == iParentB);

            Player Child1;
            Player Child2;

            int z = Random.Range(0, -10);;
            Child1 = Instantiate(playerPrefab, new Vector3(0, 1, z), Quaternion.identity);
            NeuralNetwork nnC1 = Child1.SetRandomNN();
            Child1.currentCube = -z;
            Child1.currentLane = 0;
            Child1.GetComponent <Renderer>().material.color = c;
            Child1.name = "Child1 " + Random.Range(0, 101) + "GEN" + generation;
            players.Add(Child1);


            z      = Random.Range(0, -10);;
            Child2 = Instantiate(playerPrefab, new Vector3(0, 1, z), Quaternion.identity);
            NeuralNetwork nnC2 = Child2.SetRandomNN();
            Child2.currentCube = -z;
            Child2.currentLane = 0;
            Child2.GetComponent <Renderer>().material.color = c;
            Child2.name = "Child2 " + Random.Range(0, 101) + "GEN" + generation;
            players.Add(Child2);

            NeuralNetwork nnPA = bestPlayers[iParentA].neuralNetwork;
            NeuralNetwork nnPB = bestPlayers[iParentB].neuralNetwork;

            for (int w = 0; w < nnC1.weights.Count; w++)
            {
                for (int x = 0; x < nnC1.weights[w].RowCount; x++)
                {
                    for (int y = 0; y < nnC1.weights[w].ColumnCount; y++)
                    {
                        if (Random.Range(0.0f, 1.0f) < mutateRate)
                        {
                            nnC1.weights[w][x, y] = nnPA.weights[w][x, y];
                        }

                        if (Random.Range(0.0f, 1.0f) < mutateRate)
                        {
                            nnC2.weights[w][x, y] = nnPB.weights[w][x, y];
                        }
                    }
                }
            }

            for (int w = 0; w < nnC1.biases.Count; w++)
            {
                if (Random.Range(0.0f, 1.0f) < mutateRate)
                {
                    nnC1.biases[w] = nnPB.biases[w];
                }

                if (Random.Range(0.0f, 1.0f) < mutateRate)
                {
                    nnC2.biases[w] = nnPA.biases[w];
                }
            }
        }
        else
        {
            Debug.LogError("PAS ASSEZ DE PARENTS");
        }
    }