public void Run(ref double[] input, out double[] output)
        {
            // make sure we have enough data
            if (input.Length != inputSize)
            {
                throw new ArgumentException("input data not correct dimensions");
            }

            // Dimensions
            output = new double[layerSize[layerCount - 1]];

            /* Run Network */
            for (int l = 0; l < layerCount; l++)
            {
                for (int j = 0; j < layerSize[l]; j++)
                {
                    double sum = 0.0;
                    for (int i = 0; i < (l == 0 ? inputSize : layerSize[l - 1]); i++)
                    {
                        sum += weight[l][i][j] * (l == 0 ? input[i] : layerOutput[l - 1][i]);
                    }
                    sum += biases[l][j];
                    layerInput[l][j] = sum;

                    layerOutput[l][j] = TransferFunctions.Evaluate(transferFunction[l], sum);
                }
            }
            // copy output to output array
            for (int i = 0; i < layerSize[layerCount - 1]; i++)
            {
                output[i] = layerOutput[layerCount - 1][i];
            }
        }
        public void Run(ref double[] input, out double[] output)
        {
            // Yeterli veri olup olmadığının kontrolü
            //if (input.Length != inputSize)
            //  throw new ArgumentException("Input data is not of the correct dimension.");

            // Çıktı katmanındaki nöron sayısı
            output = new double[layerSize[layerCount - 1]];

            /* Ağın çalıştırılması */
            for (int l = 0; l < layerCount; l++)
            {
                for (int j = 0; j < layerSize[l]; j++)
                {
                    double sum = 0.0;
                    for (int i = 0; i < (l == 0 ? inputSize : layerSize[l - 1]); i++)
                    {
                        sum += weight[l][i][j] * (l == 0 ? input[i] : layerOutput[l - 1][i]); //girdi değerleri ile agırlıklar çarpılıp toplanarak ilgili norondaki net toplam belirlenir
                    }
                    sum += bias[l][j];                                                        //eşik değeri bu toplama eklenir
                    layerInput[l][j] = sum;                                                   //katmanın girdisine atanır bir sonrakinde girdi olarak hesaplanması için

                    layerOutput[l][j] = TransferFunctions.Evaluate(transferFunction[l], sum); //aktivasyon fonksiyonundan geçerek noronun çıktısı belirlenir.
                }
            }

            //son(çıkış) katmanın çıktısı ağın çıktısı olarak atanır
            for (int i = 0; i < layerSize[layerCount - 1]; i++)
            {
                output[i] = layerOutput[layerCount - 1][i];
            }
        }
예제 #3
0
        public void Run(ref double[] input, out double[] output)
        {
            // Перевірка, чи введені дані відповідають кількості нейронів у вхідному шарі
            if (input.Length != inputSize)
            {
                throw new ArgumentException("Input data isn't of the correct dimension");
            }

            // Вихідне значення функції
            output = new double[layerSize[layerCount - 1]];

            // Нормалізація вхідних значень
            double max = input.Max();

            // Запуск мережі
            for (int l = 0; l < layerCount; l++)
            {
                for (int j = 0; j < layerSize[l]; j++)
                {
                    double sum = 0.0;

                    for (int i = 0; i < (l == 0 ? inputSize : layerSize[l - 1]); i++)
                    {
                        sum += weight[l][i][j] * (l == 0 ? input[i] : layerOtput[l - 1][i]);
                    }

                    sum += bias[l][j];
                    layerInput[l][j] = sum;

                    /*if (l == layerCount - 1)
                     *  layerOtput[l][j] = sum;
                     * else*/
                    layerOtput[l][j] = TransferFunctions.Evaluate(transferFunction[l], sum);
                }
            }

            // копіюємо вихід мережі у вихідний масив
            for (int i = 0; i < layerSize[layerCount - 1]; i++)
            {
                output[i] = layerOtput[layerCount - 1][i];
            }
        }
        public double Train(ref double[] input, ref double[] desired, double TrainingRate, double Momentum)
        {
            // Parameter Validation
            if (input.Length != inputSize)
            {
                throw new ArgumentException("Invalid input parameter");
            }
            if (desired.Length != layerSize[layerCount - 1])
            {
                throw new ArgumentException("Invalid input parameter");
            }

            // local variables
            double error = 0.0, sum = 0.0, weightDelta = 0.0, biasDelta = 0.0;

            double[] output = new double[layerSize[layerCount - 1]];

            // Run the Network
            Run(ref input, out output);
            // Back propagate the error
            for (int l = layerCount - 1; l >= 0; l--)
            {
                // Output layer
                if (l == layerCount - 1)
                {
                    for (int k = 0; k < layerSize[l]; k++)
                    {
                        delta[l][k]  = output[k] - desired[k];
                        error       += Math.Pow(delta[l][k], 2);
                        delta[l][k] *= TransferFunctions.EvaluateDerivative(transferFunction[l], layerInput[l][k]);
                    }
                }
                else // Hidden Layer
                {
                    for (int i = 0; i < layerSize[l]; i++)
                    {
                        sum = 0.0;
                        for (int j = 0; j < layerSize[l + 1]; j++)
                        {
                            sum += weight[l + 1][i][j] * delta[l + 1][j];
                        }
                        sum *= TransferFunctions.EvaluateDerivative(transferFunction[l], layerInput[l][i]);

                        delta[l][i] = sum;
                    }
                }
            }
            // Update the weights and biases
            for (int l = 0; l < layerCount; l++)
            {
                for (int i = 0; i < (l == 0 ? inputSize : layerSize[l - 1]); i++)
                {
                    for (int j = 0; j < layerSize[l]; j++)
                    {
                        weightDelta      = TrainingRate * delta[l][j] * (l == 0 ? input[i] : layerOutput[l - 1][i]);
                        weight[l][i][j] -= weightDelta + Momentum * previousWeight[l][i][j];

                        previousWeight[l][i][j] = weightDelta;
                    }
                }
            }

            for (int l = 0; l < layerCount; l++)
            {
                for (int i = 0; i < layerSize[l]; i++)
                {
                    biasDelta     = TrainingRate * delta[l][i];
                    biases[l][i] -= biasDelta + Momentum * previousDelta[l][i];

                    previousDelta[l][i] = biasDelta;
                }
            }
            return(error);
        }
예제 #5
0
        // Функція навчання
        public double Train(ref double[] input, ref double[] desired, double TrainingRate, double Momentum)
        {
            // Перевірка вхідних параметрів
            if (input.Length != inputSize)
            {
                throw new ArgumentException("Invalid input parameter", "input");
            }

            if (desired.Length != layerSize[layerCount - 1])
            {
                throw new ArgumentException("Invalid input parameter", "desired");
            }

            // Локальні змінні
            double error = 0.0, sum = 0.0, weigtdelta = 0.0, biasDelta = 0.0;

            double[] output = new double[layerSize[layerCount - 1]];

            // Запуск мережі
            Run(ref input, out output);

            //Розмножуємо похибку у зворотньму порядку
            for (int l = layerCount - 1; l >= 0; l--)
            {
                //Вихідний шар
                if (l == layerCount - 1)
                {
                    for (int k = 0; k < layerSize[l]; k++)
                    {
                        delta[l][k]  = output[k] - desired[k];
                        error       += Math.Pow(delta[l][k], 2);
                        delta[l][k] *= TransferFunctions.DerivativeEvaluate(transferFunction[l], layerInput[l][k]);
                    }
                }
                //Прихований шар
                else
                {
                    for (int i = 0; i < layerSize[l]; i++)
                    {
                        sum = 0.0;
                        for (int j = 0; j < layerSize[l + 1]; j++)
                        {
                            sum += weight[l + 1][i][j] * delta[l + 1][j];
                        }
                        sum        *= TransferFunctions.DerivativeEvaluate(transferFunction[l], layerInput[l][i]);
                        delta[l][i] = sum;
                    }
                }
            }

            // Оновлення ваг та відхилень
            for (int l = 0; l < layerCount; l++)
            {
                for (int i = 0; i < (l == 0 ? inputSize : layerSize[l - 1]); i++)
                {
                    for (int j = 0; j < layerSize[l]; j++)
                    {
                        weigtdelta       = TrainingRate * delta[l][j] * (l == 0 ? input[i] : layerOtput[l - 1][i]) + Momentum * previousWeightDelta[l][i][j];
                        weight[l][i][j] -= weigtdelta;

                        previousWeightDelta[l][i][j] = weigtdelta;
                    }
                }
            }

            for (int l = 0; l < layerCount; l++)
            {
                for (int i = 0; i < layerSize[l]; i++)
                {
                    biasDelta   = TrainingRate * delta[l][i] + Momentum * previosBiasDelta[l][i];
                    bias[l][i] -= biasDelta;

                    previosBiasDelta[l][i] = biasDelta;
                }
            }

            return(error);
        }
        // Функція навчання
        public double Train(ref double[] input, ref double[] desired, double kCoefficient, double bCoefficient, double eCoefficient, double Momentum)
        {
            // Перевірка вхідних параметрів
            if (input.Length != inputSize)
            {
                throw new ArgumentException("Invalid input parameter", "input");
            }

            if (desired.Length != layerSize[layerCount - 1])
            {
                throw new ArgumentException("Invalid input parameter", "desired");
            }

            // Локальні змінні
            double error = 0.0, sum = 0.0, weigtdelta = 0.0, biasDelta = 0.0, phiDelta = 0.0, phiDeltaSecond = 0.0;

            double[] output = new double[layerSize[layerCount - 1]];

            // Запуск мережі
            Run(ref input, out output);

            //Розмножуємо похибку у зворотньму порядку
            for (int l = layerCount - 1; l >= 0; l--)
            {
                //Вихідний шар
                if (l == layerCount - 1)
                {
                    for (int k = 0; k < layerSize[l]; k++)
                    {
                        delta[l][k]  = output[k] - desired[k];
                        error       += Math.Pow(delta[l][k], 2);
                        delta[l][k] *= TransferFunctions.DerivativeEvaluate(transferFunction[l], layerInput[l][k]);
                    }
                }
                //Прихований шар
                else
                {
                    for (int i = 0; i < layerSize[l]; i++)
                    {
                        sum = 0.0;
                        for (int j = 0; j < layerSize[l + 1]; j++)
                        {
                            sum += weight[l + 1][i][j] * delta[l + 1][j];
                        }
                        sum        *= TransferFunctions.DerivativeEvaluate(transferFunction[l], layerInput[l][i]);
                        delta[l][i] = sum;
                    }
                }
            }

            // Оновлення ваг та відхилень
            for (int l = 0; l < layerCount; l++)
            {
                for (int i = 0; i < (l == 0 ? inputSize : layerSize[l - 1]); i++)
                {
                    for (int j = 0; j < layerSize[l]; j++)
                    {
                        phiDelta       = delta[l][j] * (l == 0 ? input[i] : layerOtput[l - 1][i]); // обрахунок phi для поточної ваги
                        phiDeltaSecond = (1 - eCoefficient) * previousPhiDelta[l][i][j]
                                         + eCoefficient * previousPhiDeltaSecond[l][i][j];         // обрахунок phiSecond для поточної ваги

                        // оновлення швидкості навчання для ваги
                        learningRates[l][i][j] +=
                            CalculateLearningRateDelta(previousPhiDeltaSecond[l][i][j], phiDelta, kCoefficient, bCoefficient, learningRates[l][i][j]);

                        weigtdelta       = learningRates[l][i][j] * phiDelta + Momentum * previousWeightDelta[l][i][j];
                        weight[l][i][j] -= weigtdelta;

                        previousWeightDelta[l][i][j]    = weigtdelta;
                        previousPhiDelta[l][i][j]       = phiDelta;
                        previousPhiDeltaSecond[l][i][j] = phiDeltaSecond;
                    }
                }
            }

            for (int l = 0; l < layerCount; l++)
            {
                for (int i = 0; i < layerSize[l]; i++)
                {
                    biasDelta   = initialLearningRate * delta[l][i] + Momentum * previosBiasDelta[l][i];
                    bias[l][i] -= biasDelta;

                    previosBiasDelta[l][i] = biasDelta;
                }
            }

            return(error);
        }
        //ağın eğitilmesi backpropagation
        public double Train(ref double[] input, ref double[] desired, double TrainingRate, double Momentum)
        {
            // Local variable
            double error = 0.0, sum = 0.0, weightDelta = 0.0, biasDelta = 0.0;

            double[] output = new double[layerSize[layerCount - 1]];

            // Ağın çalıştırılması
            Run(ref input, out output);

            // Hatayı geri yaymak
            for (int l = layerCount - 1; l >= 0; l--)
            {
                // Çıktı katmanı
                if (l == layerCount - 1)
                {
                    for (int k = 0; k < layerSize[l]; k++)
                    {
                        delta[l][k]  = output[k] - desired[k];
                        error       += Math.Pow(delta[l][k], 2);
                        delta[l][k] *= TransferFunctions.EvaluateDerivative(transferFunction[l],
                                                                            layerInput[l][k]);
                    }
                }
                else // Gizli katman
                {
                    for (int i = 0; i < layerSize[l]; i++)
                    {
                        sum = 0.0;
                        for (int j = 0; j < layerSize[l + 1]; j++)
                        {
                            sum += weight[l + 1][i][j] * delta[l + 1][j];
                        }
                        sum *= TransferFunctions.EvaluateDerivative(transferFunction[l], layerInput[l][i]);

                        delta[l][i] = sum;
                    }
                }
            }

            // Ağırlıklar ve bias değerlerinin güncellenmesi
            for (int l = 0; l < layerCount; l++)
            {
                for (int i = 0; i < (l == 0 ? inputSize : layerSize[l - 1]); i++)
                {
                    for (int j = 0; j < layerSize[l]; j++)
                    {
                        weightDelta = TrainingRate * delta[l][j] * (l == 0 ? input[i] : layerOutput[l - 1][i])
                                      + Momentum * previousWeightDelta[l][i][j];
                        weight[l][i][j] -= weightDelta;

                        previousWeightDelta[l][i][j] = weightDelta;
                    }
                }
            }

            for (int l = 0; l < layerCount; l++)
            {
                for (int i = 0; i < layerSize[l]; i++)
                {
                    biasDelta   = TrainingRate * delta[l][i];
                    bias[l][i] -= biasDelta + Momentum * previousBiasDelta[l][i];

                    previousBiasDelta[l][i] = biasDelta;
                }
            }

            return(error);
        }