void TrainNetwork(int itterations)
{
for (int i = 0; i < itterations; i++)
{
int random = Random.Range(0, trainingData.Length);
float[] generatorInput =
{
trainingData[random].targetResult[0],
trainingData[random].targetResult[1],
trainingData[random].targetResult[2],
trainingData[random].targetResult[3],
trainingData[random].targetResult[4],
trainingData[random].targetResult[5],
trainingData[random].targetResult[6],
trainingData[random].targetResult[7],
trainingData[random].targetResult[8],
trainingData[random].targetResult[9],
RandomGaussian()
};
float[] generatedImage = generator.Guess(generatorInput);
float[] realOutput = discriminator.Guess(trainingData[random].input);
float[] fakeOutput = discriminator.Guess(generatedImage);
float discriminatorLoss = CalculateLoss(realOutput);
float generatorLoss = CalculateLoss(fakeOutput);
}
}
public void TestNeuralNetwork()
{
int[] fails = new int[10];
int failsTotal = 0;
for (int i = 0; i < testImages.imageCount; i++)
{
float[] input = new float[testImages.pixelValueArrayOffset];
for (int j = 0; j < testImages.pixelValueArrayOffset; j++)
{
input[j] =
testImages.pixelValueArray[i * testImages.pixelValueArrayOffset + j];
}
float[] result = neuralNetwork.Guess(input);
if (!IsResultCorrect(result, testImages.labelValueArray[i]))
{
fails[testImages.labelValueArray[i]]++;
failsTotal++;
}
}
Debug.Log($"{failsTotal} fails:");
Debug.Log((1 - (float)failsTotal / (float)(testImages.imageCount)) * 100 + " % accurate");
}
public void TestNeuralNetwork(int itterations)
{
int failsTotal = 0;
for (int i = 0; i < itterations; i++)
{
int random = UnityEngine.Random.Range(0, trainingData.Length);
float[] result = neuralNetwork.Guess(trainingData[random].input);
if (!IsResultCorrect(result, trainingData[random].targetResult))
{
failsTotal++;
}
}
Debug.Log($"{failsTotal} fails:");
Debug.Log((1 - (float)failsTotal / (float)(itterations)) * 100 + " % accurate");
}
public void CreateImange()
{
GeneratedNumber.text = targetNumber.ToString();
float[] input = new float[11];
input[targetNumber] = 1;
input[10] = slider.value;
float[] pixelValues = neuralNetwork.Guess(input);
for (int i = 0; i < 28; i++)
{
for (int j = 0; j < 28; j++)
{
image.sprite.texture.SetPixel(i, j, new Color(pixelValues[i + j * 28], 0, 0));
}
}
image.sprite.texture.Apply();
}
// Start is called before the first frame update
void Start()
{
TrainingData[] trainingData = new TrainingData[dataSet.Length];
for (int i = 0; i < trainingData.Length; i++)
{
trainingData[i] = new TrainingData
{
input = new float[] {
dataSet[i].sepal_length,
dataSet[i].sepal_width,
dataSet[i].petal_length,
dataSet[i].sepal_width
},
targetResult = dataSet[i].species,
};
}
//Normalize(trainingData);
NeuralNetwork_Matrix neuralNetwork = new NeuralNetwork_Matrix(
trainingData[0].input.Length, 7, trainingData[0].targetResult.Length);
neuralNetwork.TrainNeuralNetwork(trainingData, 10_000, true);
int itterations = 100;
int fails = 0;
for (int i = 0; i < itterations; i++)
{
int random = Random.Range(0, trainingData.Length);
float[] result = neuralNetwork.Guess(trainingData[random].input);
Debug.Log($"{result[0]} {result[1]} {result[2]} |||| {trainingData[random].targetResult[0]} {trainingData[random].targetResult[1]} {trainingData[random].targetResult[2]}");
if (!IsResultCorrect(result, trainingData[random].targetResult))
{
fails++;
}
}
Debug.Log((1 - (float)fails / (float)itterations) * 100 + " % accurate");
}
