/// <summary>
/// Обучить нейронную сеть на накопленных исследованиях
/// </summary>
/// <param name="countLifes">На скольких жизней, начиная от последней, обучить модель</param>
/// <param name="epochs">Количество эпох обучения. По умолчанию 1</param>
/// <param name="learningRate">Норма обучения. По умолчанию 1e-3</param>
/// <param name="trainType">Тип обучения. По умолчанию online</param>
/// <param name="minLoss">ошибка, при которой обучение останавливается</param>
/// <param name="optimizer">Оптимизатор. По умолчанию Adam</param>
/// <param name="loss">Метрика ошибки. По умолчанию MSE</param>
public void Train(int countLifes = 50, int epochs = 1, double learningRate = 1e-3, TrainType trainType = TrainType.Online, double minLoss = 0.0, IOptimizer optimizer = null, ILoss loss = null)
{
if (loss == null)
{
loss = new LossMeanSqrSqrt();
}
if (optimizer == null)
{
optimizer = new Adam();
}
int start = lifes.Count - countLifes;
Vector rewards = GetRewards(start, lifes.Count);
var inputs = new List <NNValue>();
var outputs = new List <NNValue>();
for (int i = 0; i < rewards.N; i++)
{
var conditions = lifes[start + i].GetConditions();
foreach (var condition in conditions)
{
var state = condition.Item1;
var action = condition.Item2;
inputs.Add(state.ToNNValue());
if (rewards[i] > 0)
{
outputs.Add(new NNValue(action.probabilities.MaxOutVector().TransformVector(x => (x == -1) ? 0 : 1)));
}
else
{
outputs.Add(new NNValue((1.0 - action.probabilities).MaxOutVector().TransformVector(x => (x == -1) ? 0 : 1)));
}
}
}
#region Shuffle
for (int i = start; i < inputs.Count; i++)
{
var a = random.Next(start, inputs.Count);
var b = random.Next(start, inputs.Count);
var temp1 = inputs[a];
var temp2 = outputs[a];
inputs[a] = inputs[b];
outputs[a] = outputs[b];
inputs[b] = temp1;
outputs[b] = temp2;
}
#endregion
#region Train
DataSetNoReccurent dataSetNoReccurent = new DataSetNoReccurent(inputs.ToArray(), outputs.ToArray(), loss);
Trainer trainer = new Trainer(graphBackward, trainType, optimizer);
trainer.Train(epochs, learningRate, model, dataSetNoReccurent, minLoss);
#endregion
}
// Обучение нейронной сети
private void button2_Click(object sender, EventArgs e)
{
GetData();
DataSetNoReccurent DatasetNo = new DataSetNoReccurent(x, y, new CrossEntropy(), 0.4);
Trainer trainer = new Trainer(new GraphCPU(), TrainType.Online, new Adam());
trainer.Train(8, 0.001, NNW, DatasetNo, 0.0006);
}
static void Main(string[] args)
{
Random random = new Random(13);
NeuralNetwork cNN = new NeuralNetwork(random, 0.2);
var conv = new ConvolutionLayer(new RectifiedLinearUnit(0.01), 8, 3, 3);
conv.IsSame = true;
cNN.AddNewLayer(new Shape(28, 28), conv);
cNN.AddNewLayer(new MaxPooling(2, 2));
cNN.AddNewLayer(new ConvolutionLayer(new RectifiedLinearUnit(0.01), 16, 3, 3));
cNN.AddNewLayer(new MaxPooling(2, 2));
cNN.AddNewLayer(new ConvolutionLayer(new RectifiedLinearUnit(0.01), 32, 3, 3));
cNN.AddNewLayer(new UnPooling(2, 2));
cNN.AddNewLayer(new ConvolutionLayer(new RectifiedLinearUnit(0.01), 16, 3, 3));
cNN.AddNewLayer(new MaxPooling(2, 2));
cNN.AddNewLayer(new Flatten());
cNN.AddNewLayer(new FeedForwardLayer(20, new RectifiedLinearUnit(0.01)));
cNN.AddNewLayer(new FeedForwardLayer(2, new SoftmaxUnit()));
Console.WriteLine(cNN);
GraphCPU graph = new GraphCPU(false);
NNValue nValue = NNValue.Random(28, 28, 2, random);
NNValue nValue1 = NNValue.Random(28, 28, 2, random);
NNValue outp = new NNValue(new double[] { 0, 1 });
NNValue outp1 = new NNValue(new double[] { 1, 0 });
DataSetNoReccurent data = new DataSetNoReccurent(new NNValue[] { nValue, nValue1 }, new NNValue[] { outp, outp1 }, new CrossEntropyWithSoftmax());
TrainerCPU trainer = new TrainerCPU(TrainType.MiniBatch, new Adam());
trainer.BatchSize = 2;
trainer.Train(10000, 0.001, cNN, data, 2, 0.0001);
double[] dbs = cNN.Activate(nValue, graph).DataInTensor;
double[] dbs1 = cNN.Activate(nValue1, graph).DataInTensor;
}
/// <summary>
/// Обучить нейронную сеть на накопленных исследованиях
/// </summary>
/// <param name="countGenerations">На скольких жизней, начиная от последней, обучить модель</param>
/// <param name="epochs">Количество эпох обучения. По умолчанию 1</param>
/// <param name="learningRate">Норма обучения. По умолчанию 1e-3</param>
/// <param name="trainType">Тип обучения. По умолчанию online</param>
/// <param name="minLoss">ошибка, при которой обучение останавливается</param>
/// <param name="optimizer">Оптимизатор. По умолчанию Adam</param>
/// <param name="loss">Метрика ошибки. По умолчанию MSE</param>
public void Train(int countGenerations = -1, int epochs = 1, float learningRate = 1e-3f, TrainType trainType = TrainType.Online, float minLoss = 0.0f, ILoss loss = null)
{
if (loss == null)
{
loss = new LossMeanSqrSqrt();
}
if (countGenerations == -1)
{
countGenerations = Generations.Count;
}
int start = Generations.Count - countGenerations;
int end = start + countGenerations;
if (end > Generations.Count)
{
end = Generations.Count;
}
Vector rewards = GetRewards(start, end);
var inputs = new List <NNValue>();
var outputs = new List <NNValue>();
for (int i = 0; i < rewards.Count; i++)
{
var conditions = Generations[start + i].GetConditions();
foreach (var condition in conditions)
{
var state = condition.Item1;
var action = condition.Item2;
double p = 0.01;
if (rewards[i] > 0)
{
inputs.Add(state.Input);
Vector outp = new Vector(degreesOfFreedom);
outp[action.index] = 1.0 - p;
outputs.Add(new NNValue(outp));
}
else if (rewards[i] < 0)
{
inputs.Add(state.Input);
int u = 0;
while ((u = random.Next(0, degreesOfFreedom)) == action.index)
{
;
}
Vector output = new Vector(degreesOfFreedom);
output[u] = 1.0 - p;
outputs.Add(new NNValue(output));
}
}
}
if (inputs.Count > 0)
{
Shuffle(inputs, outputs);
#region Train
DataSetNoReccurent dataSetNoReccurent = new DataSetNoReccurent(inputs.ToArray(), outputs.ToArray(), loss);
trainer.Train(epochs, learningRate, model, dataSetNoReccurent, minLoss);
#endregion
}
}
