void CreateLinearMulticlass()
{
for (int i = 0; i < _outputSize; i++)
{
_myMulticlassModel.Add(MlDllWrapper.CreateLinearModel(_inputSize));
}
}
void trainLinearMulticlass()
{
print("Starting trainning");
_outputTrainningMulticlass.Clear();
_outputTrainningMulticlass = new List <double[]>();
int outputLength = trainningOuput.Length / _outputSize;
for (int i = 0; i < _outputSize; i++)
{
List <double> tmp = new List <double>();
for (int j = 0; j < outputLength; j++)
{
tmp.Add(trainningOuput[j * _outputSize + i]);
}
_outputTrainningMulticlass.Add(tmp.ToArray());
}
for (int i = 0; i < _outputTrainningMulticlass.Count; i++)
{
MlDllWrapper.trainLinearClass(_myMulticlassModel[i], trainningInput, trainningInput.Length, _inputSize,
_outputTrainningMulticlass[i],
_outputTrainningMulticlass[i].Length, 1, learningRate, isClassification, epochs);
}
print("Trainning Finished");
}
void Update()
{
if (Input.GetKeyDown(KeyCode.A))
{
if (!isClassification)
{
predictMLPMulticlassRegression();
}
else
{
predictMLPMulticlass();
}
}
if (Input.GetKeyDown(KeyCode.T))
{
print("Starting trainning");
MlDllWrapper.trainMLPModelClass(MyModel, numberLayer, trainningInput.Length / npl[0], npl,
trainningInput,
trainningInput.Length, _inputSize, trainningOuput, trainningOuput.Length, _outputSize, epochs,
learningRate, isClassification);
print("Trainning Finished");
//trainModel();
//trainLinearMulticlass();
}
if (Input.GetKeyDown(KeyCode.R))
{
MlDllWrapper.DeleteLinearModel(MyModel);
}
}
void Update()
{
if (Input.GetKeyDown(KeyCode.A))
{
predictMLPMulticlass();
}
if (Input.GetKeyDown(KeyCode.I))
{
File.Copy(userRequestPathFile, @".\Assets\Resources\Dataset\requestImage.jpg", true);
requestPicRead();
}
if (Input.GetKeyDown(KeyCode.T))
{
print("Starting trainning");
MlDllWrapper.trainMLPModelClass(MyModel, numberLayer, trainningInput.Length / npl[0], npl, trainningInput,
trainningInput.Length, _inputSize, trainningOuput, trainningOuput.Length, _outputSize, epochs,
learningRate, isClassification);
evaluateDataset();
print("Trainning Finished");
}
if (Input.GetKeyDown(KeyCode.R))
{
MlDllWrapper.DeleteLinearModel(MyModel);
}
if (Input.GetKeyDown(KeyCode.O))
{
StopAllCoroutines();
StartCoroutine(infiniteTrain());
}
}
void predictMLPMulticlass()
{
int bestModel = 0;
double bestValue = 0.0f;
IntPtr pointerToValues = MlDllWrapper.PredictMLPModel(MyModel, requestInput, _inputSize,
numberLayer, npl, isClassification);
string[] arrayValuesasString = Marshal.PtrToStringAnsi(pointerToValues).Split(';');
for (int i = 0; i < _outputSize; i++)
{
double currentValue = Convert.ToDouble(arrayValuesasString[i].Replace('.', ','));
print($"{currentValue} and model {i}");
//print($"{currentValue} and model {i}");
if (currentValue > bestValue)
{
bestModel = i;
bestValue = currentValue;
}
}
print($"The best value is {bestValue} with the model {bestModel}");
//Regression Lineaire
/*if(pointerToValues != IntPtr.Zero)
* MlDllWrapper.DeleteLinearModel(pointerToValues);*/
}
void trainModel()
{
print("Model start trainning");
MlDllWrapper.trainLinearClass(MyModel, trainningInput, trainningInput.Length, 2,
trainningOuput, trainningOuput.Length, 1, learningRate, true, epochs);
print("Model finished train");
}
void Start()
{
CsvReader.readCSVFile("TrainningData\\" + trainningFile, ref inputList, ref outputList);
trainningInput = inputList.ToArray();
trainningOuput = outputList.ToArray();
// MlDllWrapper.InitRBF(trainningInput, trainningInput.Length, 784, trainningOuput,
// trainningOuput.Length, 10, 500);
_inputSize = npl[0];
_outputSize = npl.Last();
numberLayer = npl.Length;
// //
if (CsvReader.inputCount != _inputSize)
{
Debug.LogError(
$"Input Length ({CsvReader.inputCount}) in CSV File don't match the npl input length ({_inputSize})");
}
if (CsvReader.outputCount != _outputSize)
{
Debug.LogError(
$"Output Length ({CsvReader.outputCount}) in CSV File don't match the npl output length ({_outputSize})");
}
//MyModel = MlDllWrapper.CreateLinearModel(modelSize);
// //CreateLinearMulticlass();
//
//
if (isRandomized)
{
randomizeSpheres(minRandom, maxRandom);
}
//
testDataSet = new List <double[]>();
// //
foreach (sphereExposer sphere in spheres)
{
Vector3 pos = sphere.myTransform.position;
double[] tmp = new[] { (double)pos.x, pos.y };
testDataSet.Add(tmp);
// }
// //
// // //trainModel();
// //
// // //predictOnDataSet();
// //
MyModel = MlDllWrapper.CreateMLPModel(numberLayer, npl);
}
// Update is called once per frame
}
IEnumerator infiniteTrain()
{
int realEpochs = 0;
while (true)
{
MlDllWrapper.trainMLPModelClass(MyModel, numberLayer, trainningInput.Length / npl[0], npl, trainningInput,
trainningInput.Length, _inputSize, trainningOuput, trainningOuput.Length, _outputSize, epochs,
learningRate, isClassification);
evaluateDataset();
realEpochs++;
yield return(null);
}
}
// Start is called before the first frame update
void Start()
{
CsvReader.readCSVFile("TrainningData\\inputCsv.csv", ref inputList, ref outputList);
trainningInput = inputList.ToArray();
trainningOuput = outputList.ToArray();
_inputSize = npl[0];
_outputSize = npl.Last();
numberLayer = npl.Length;
if (CsvReader.inputCount != _inputSize)
{
Debug.LogError($"Input Length ({CsvReader.inputCount}) in CSV File don't match the npl input length ({_inputSize})");
}
if (CsvReader.outputCount != _outputSize)
{
Debug.LogError($"Output Length ({CsvReader.outputCount}) in CSV File don't match the npl output length ({_outputSize})");
}
MyModel = MlDllWrapper.CreateMLPModel(numberLayer, npl);
}
void predictLinearMulticlass()
{
for (int j = 0; j < testDataSet.Count; j++)
{
_multiclassResult.Clear();
int bestModel = 0;
double bestValue = 0.0f;
for (int i = 0; i < _outputSize; i++)
{
double currentValue = MlDllWrapper.PredictLinearModel(_myMulticlassModel[i], testDataSet[j], _inputSize,
isClassification);
print($"Model numero {i} = {currentValue}");
if (currentValue > bestValue)
{
bestModel = i;
bestValue = currentValue;
}
}
switch (bestModel)
{
default:
spheres[j].ChangeMaterial(blueMat);
break;
case 1:
spheres[j].ChangeMaterial(redMat);
break;
case 2:
spheres[j].ChangeMaterial(greenMat);
break;
}
}
}
void predictMLPMulticlassRegression()
{
for (int j = 0; j < testDataSet.Count; j++)
{
_multiclassResult.Clear();
int bestModel = 0;
double bestValue = 0.0f;
IntPtr pointerToValues = MlDllWrapper.PredictMLPModel(MyModel, testDataSet[j], _inputSize,
numberLayer, npl, isClassification);
string[] arrayValuesasString = Marshal.PtrToStringAnsi(pointerToValues).Split(';');
for (int i = 0; i < _outputSize; i++)
{
double currentValue = Convert.ToDouble(arrayValuesasString[i].Replace('.', ','));
print($"{currentValue} and model {i}");
if (currentValue > bestValue)
{
bestModel = i;
bestValue = currentValue;
}
}
print(arrayValuesasString[0]);
//Regression Lineaire
spheres[j].changeZ((float)Convert.ToDouble(arrayValuesasString[0].Replace('.', ',')));
/*if(pointerToValues != IntPtr.Zero)
* MlDllWrapper.DeleteLinearModel(pointerToValues);*/
}
}
void predictMLPMulticlass()
{
for (int j = 0; j < testDataSet.Count; j++)
{
_multiclassResult.Clear();
int bestModel = 0;
double bestValue = 0.0f;
IntPtr pointerToValues = MlDllWrapper.PredictMLPModel(MyModel, testDataSet[j], _inputSize,
numberLayer, npl, isClassification);
string[] arrayValuesasString = Marshal.PtrToStringAnsi(pointerToValues).Split(';');
for (int i = 0; i < _outputSize; i++)
{
double currentValue = Convert.ToDouble(arrayValuesasString[i].Replace('.', ','));
if (currentValue > 0)
{
spheres[j].ChangeMaterial(blueMat);
}
else
{
spheres[j].ChangeMaterial(redMat);
}
}
/*if(pointerToValues != IntPtr.Zero)
* MlDllWrapper.DeleteLinearModel(pointerToValues);*/
}
}
int predictDatasetSample(int k)
{
int bestModel = 0;
double bestValue = 0.0f;
List <double> tmp = new List <double>();
for (int i = k * npl[0]; i < npl[0] * (k + 1); i++)
{
tmp.Add(trainningInput[i]);
}
double[] tmp2 = tmp.ToArray();
IntPtr pointerToValues = MlDllWrapper.PredictMLPModel(MyModel, tmp2, _inputSize,
numberLayer, npl, isClassification);
string[] arrayValuesasString = Marshal.PtrToStringAnsi(pointerToValues).Split(';');
for (int i = 0; i < _outputSize; i++)
{
double currentValue = Convert.ToDouble(arrayValuesasString[i].Replace('.', ','));
// print($"{currentValue} and model {i}");
//print($"{currentValue} and model {i}");
if (currentValue > bestValue)
{
bestModel = i;
bestValue = currentValue;
}
}
return(bestModel);
}
