static void Real()
{
var inputLayer = new InputLayer3D(1, 1, 1);
var outputLayer = new OutputLayer(1)
{
ActivationFunction = new ConstOutputArrayFunction()
};
var dataProvider = new FunctionProvider();
InitializeTrainingData(dataProvider.TrainData);
var perceptron1 = new PerceptronLayer(10, 2)
{
ActivationFunction = new SigmoidFunction()
};
var perceptron2 = new PerceptronLayer(10, 10)
{
ActivationFunction = new SigmoidFunction()
};
var perceptron3 = new PerceptronLayer(8, 10)
{
ActivationFunction = new SigmoidFunction()
};
var perceptron4 = new PerceptronLayer(6, 8)
{
ActivationFunction = new SigmoidFunction()
};
var perceptron5 = new PerceptronLayer(2, 6)
{
ActivationFunction = new SigmoidFunction()
};
MultiLayerPerceptron network = new MultiLayerPerceptron
{
InputLayer = inputLayer,
OutputLayer = outputLayer,
DataProvider = dataProvider
};
network.HiddenLayers.Add(perceptron1);
network.HiddenLayers.Add(perceptron2);
network.HiddenLayers.Add(perceptron3);
network.HiddenLayers.Add(perceptron4);
network.HiddenLayers.Add(perceptron5);
var trainer = new FCTrainer(network, 10, 1, dataProvider);
trainer.Train(1);
var error = network.Test(1);
}
static void OneTrainingData()
{
var inputLayer = new InputLayer3D(1, 1, 3);
var outputLayer = new OutputLayer(2)
{
ActivationFunction = new ConstOutputArrayFunction()
};
var weight1 = new List <Array3D> {
new Array3D(0.1, 0.3), new Array3D(0.3, 0.1)
};
var weight2 = new List <Array3D> {
new Array3D(0.4, 0.5), new Array3D(0.3, 0.5)
};
var perceptron1 = new PerceptronLayer(weight1)
{
ActivationFunction = new SigmoidFunction()
};
var perceptron2 = new PerceptronLayer(weight2)
{
ActivationFunction = new SigmoidFunction()
};
var dataProvider = new FunctionProvider
{
TrainData =
{
new TrainingData <Array3D, Array3D> {
Input = new Array3D(0.3, 0.4, 0.5), Expected = new Array3D(0.2, 0.6)
},
new TrainingData <Array3D, Array3D> {
Input = new Array3D(0.2, 0.4, 0.7), Expected = new Array3D(0.1, 0.8)
}
}
};
var network = new MultiLayerPerceptron
{
InputLayer = inputLayer,
OutputLayer = outputLayer,
DataProvider = dataProvider
};
network.HiddenLayers.Add(perceptron1);
network.HiddenLayers.Add(perceptron2);
var trainer = new FCTrainer(network, 2, 1, dataProvider);
trainer.Train(100);
}
static void Xor()
{
const int batchSize = 4;
const int epochSize = 16;
var inputLayer = new InputLayer3D(1, 1, 1);
var outputLayer = new OutputLayer(1)
{
ActivationFunction = new ConstOutputArrayFunction()
};
var dataProvider = new FunctionProvider
{
TrainData =
{
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 0), Expected = new Array3D(0.0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 1), Expected = new Array3D(1.0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 0), Expected = new Array3D(1.0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 1), Expected = new Array3D(0.0)
}
},
TestData = { new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 0), Expected = new Array3D(0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 1), Expected = new Array3D(1)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 0), Expected = new Array3D(1)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 1), Expected = new Array3D(0)
} },
IsQueue = false
};
var oneData = new FunctionProvider {
TrainData =
{
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 0), Expected = new Array3D(0.0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 1), Expected = new Array3D(0.0)
}
}, IsQueue = false
};
var function = new FunctionProvider(x => Math.Pow(x, 2));
var weight1 = new List <Array3D> {
new Array3D(0.1, 0.3), new Array3D(0.3, 0.1)
};
var weight2 = new List <Array3D> {
new Array3D(0.4, 0.5)
};
var perceptron1 = new PerceptronLayer(5, 2)
{
ActivationFunction = new TanhActivationFunction()
};
perceptron1.Trainer = new MiniBatchPerceptronTrainer(perceptron1.Neurals, false)
{
BatchSize = batchSize, ActivationFunction = new TanhActivationFunction(), LearningRate = 0.1, Momentum = 0.1
};
var perceptron2 = new PerceptronLayer(1, 5)
{
ActivationFunction = new TanhActivationFunction()
};
perceptron2.Trainer = new MiniBatchPerceptronTrainer(perceptron2.Neurals, true)
{
BatchSize = batchSize, ActivationFunction = new TanhActivationFunction(), LearningRate = 0.1, Momentum = 0.1
};
var network = new MultiLayerPerceptron
{
InputLayer = inputLayer,
OutputLayer = outputLayer,
DataProvider = dataProvider
};
network.HiddenLayers.Add(perceptron1);
network.HiddenLayers.Add(perceptron2);
var trainer = new FCTrainer(network, epochSize, batchSize, dataProvider);
trainer.Train(200);
}
public void TestNetwork()
{
const int layer0 = 3, layer1 = 2, layer2 = 3;
#region 1 Iteration
double[] input = { 0.8, 0.76, 0.54 };
double[] weight1 = { 0.35, 0.46, 0.51 };
double[] weight2 = { 0.4, 0.87, 0.36 };
double[] sum1 = { 0, 0 };
for (var i = 0; i < input.Length; i++)
{
sum1[0] += input[i] * weight1[i];
sum1[1] += input[i] * weight2[i];
}
double[] output1 = { Expanent(sum1[0]), Expanent(sum1[1]) };
#endregion
#region 2 Iteration
double[] weight3 = { 0.5, 0.14 };
double[] weight4 = { 0.76, 0.86 };
double[] weight5 = { 0.95, 0.12 };
double[] sum2 = { 0, 0, 0 };
for (var i = 0; i < output1.Length; i++)
{
sum2[0] += output1[i] * weight3[i];
sum2[1] += output1[i] * weight4[i];
sum2[2] += output1[i] * weight5[i];
}
double[] output2 = { Expanent(sum2[0]), Expanent(sum2[1]), Expanent(sum2[2]) };
#endregion
#region Train
double[] expected = { 0.87, 1, 0.32 };
double[] error = new double[layer2];
for (var i = 0; i < error.Length; i++)
{
error[i] = output2[i] - expected[i];
}
double[] weightDelta = new double[layer2];
for (var i = 0; i < layer2; i++)
{
weightDelta[i] = error[i] * output2[i] * (1 - output2[i]);
}
var newWeight3 = new double[layer1];
var newWeight4 = new double[layer1];
var newWeight5 = new double[layer1];
const double learningRate = 0.1;
for (var i = 0; i < layer1; i++)
{
newWeight3[i] = weight3[i] - output1[i] * weightDelta[0] * learningRate;
newWeight4[i] = weight4[i] - output1[i] * weightDelta[1] * learningRate;
newWeight5[i] = weight5[i] - output1[i] * weightDelta[2] * learningRate;
}
#region 2 Iteration
double[] error2 = new double[layer1];
for (var i = 0; i < layer1; i++)
{
error2[i] += weight3[i] * weightDelta[0];
error2[i] += weight4[i] * weightDelta[1];
error2[i] += weight5[i] * weightDelta[2];
}
double[] weightDelta2 = new double[layer1];
for (var i = 0; i < layer1; i++)
{
weightDelta2[i] = error2[i] * output1[i] * (1 - output1[i]);
}
var newWeight1 = new double[layer0];
var newWeight2 = new double[layer0];
for (var i = 0; i < layer0; i++)
{
newWeight1[i] = weight1[i] - input[i] * weightDelta2[0] * learningRate;
newWeight2[i] = weight2[i] - input[i] * weightDelta2[1] * learningRate;
}
#endregion
#endregion
#region Compare
var inputLayer = new InputLayer3D(1, 1, 3);
var outputLayer = new OutputLayer(3)
{
ActivationFunction = new ConstArrayFunction()
};
var weightList1 = new List <List <double> > {
new List <double>(weight1), new List <double>(weight2)
};
var hiddenLayer1 = new PerceptronLayer(weightList1);
var weightList2 = new List <List <double> >
{
new List <double>(weight3),
new List <double>(weight4),
new List <double>(weight5)
};
var hiddenLayer2 = new PerceptronLayer(weightList2);
var trainingData = new TrainingData <Array3D, Array3D>
{
Input = new Array3D(input[0], input[1], input[2]),
Expected = new Array3D(expected[0], expected[1], expected[2])
};
var trainingData2 = new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0.15, 0.44, 0.83),
Expected = new Array3D(0.1, 0.24, 0.18)
};
var network = new MultiLayerPerceptron
{
InputLayer = inputLayer,
OutputLayer = outputLayer,
HiddenLayers = { hiddenLayer1, hiddenLayer2 },
DataProvider = new FunctionProvider {
TrainData = { trainingData, trainingData2 }, IsQueue = true
}
};
for (var i = 0; i < 3; i++)
{
Assert.AreEqual(network.HiddenLayers[0].Neurals[0].Weights[i], newWeight1[i], 0.001);
Assert.AreEqual(network.HiddenLayers[0].Neurals[1].Weights[i], newWeight2[i], 0.001);
}
#endregion
}
private void InitializeNetwork()
{
var inputLayer = new InputLayer3D(1, 1, 1);
var outputLayer = new OutputLayer(1)
{
ActivationFunction = new ConstOutputArrayFunction()
};
var dataProvider = new FunctionProvider
{
TrainData =
{
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 0), Expected = new Array3D(0.0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 1), Expected = new Array3D(1.0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 0), Expected = new Array3D(1.0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 1), Expected = new Array3D(0.0)
}
},
TestData = { new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 0), Expected = new Array3D(0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 1), Expected = new Array3D(1)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 0), Expected = new Array3D(1)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 1), Expected = new Array3D(0)
} },
IsQueue = false
};
var oneData = new FunctionProvider
{
TrainData =
{
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(0, 0), Expected = new Array3D(0.0)
},
new TrainingData <Array3D, Array3D>
{
Input = new Array3D(1, 1), Expected = new Array3D(0.0)
}
},
IsQueue = false
};
var function = new FunctionProvider(x => Math.Pow(x, 2));
var perceptron1 = new PerceptronLayer(10, 1)
{
ActivationFunction = new TanhActivationFunction()
};
perceptron1.Trainer = new MiniBatchPerceptronTrainer(perceptron1.Neurals, false)
{
BatchSize = BatchSize, ActivationFunction = new TanhActivationFunction(), LearningRate = 0.1, Momentum = 0.1
};
var perceptron2 = new PerceptronLayer(1, 10)
{
ActivationFunction = new TanhActivationFunction()
};
perceptron2.Trainer = new MiniBatchPerceptronTrainer(perceptron2.Neurals, true)
{
BatchSize = BatchSize, ActivationFunction = new TanhActivationFunction(), LearningRate = 0.1, Momentum = 0.1
};
Network = new MultiLayerPerceptron
{
InputLayer = inputLayer,
OutputLayer = outputLayer,
DataProvider = function
};
Network.HiddenLayers.Add(perceptron1);
Network.HiddenLayers.Add(perceptron2);
Trainer = new FCTrainer(Network, EpochSize, BatchSize, function);
}
