IntPtr CreateModel(int size)
{
Debug.Log("Create Model");
var modelPtr = VisualStudioLibWrapper.linear_model_create(size);
/*double[] model = new double[size+1];
* Marshal.Copy(modelPtr, model, 0, size + 1);
* Debug.Log(model[size]);*/
return(modelPtr);
}
public unsafe void TrainAndTest()
{
Debug.Log("Training and Testing");
var Y = new double[trainSpheresTransforms.Length];
IntPtr model;
var linear_inputs = new double[trainSpheresTransforms.Length * 2];
for (int i = 0; i < trainSpheresTransforms.Length; i++)
{
var position = new Vector3();
if (transfo != -1)
{
position = AppliTransfo(trainSpheresTransforms[i].position);
}
else
{
position = trainSpheresTransforms[i].position;
}
Y[i] = position.y;
linear_inputs[i * 2] = position.x;
linear_inputs[(i * 2) + 1] = position.z;
}
// Create Model
model = VisualStudioLibWrapper.linear_model_create(2);
// Train Model
VisualStudioLibWrapper.linear_model_train_regression(model, linear_inputs, trainSpheresTransforms.Length, 2, Y, Y.Length, 1000000, 0.01f);
// For each testSphere : Predict
foreach (var testSpheres in testSpheresTransforms)
{
var position = new Vector3();
if (transfo != -1)
{
position = AppliTransfo(testSpheres.position);
}
else
{
position = testSpheres.position;
}
double[] input = { position.x, position.z };
float y = (float)VisualStudioLibWrapper.linear_model_predict_regression(model, input, 2);
testSpheres.position = new Vector3(
position.x,
y,
position.z
);
}
}
public unsafe void TrainAndTest()
{
Debug.Log("Training and Testing");
int count = 0;
int blue_count = 0;
int red_count = 0;
// Créer dataset_inputs
// Créer dataest_expected_outputs
var Y = new double[trainSpheresTransforms.Length];
IntPtr model;
var linear_inputs = new double[trainSpheresTransforms.Length * 2];
for (int i = 0; i < trainSpheresTransforms.Length; i++)
{
Y[i] = trainSpheresTransforms[i].position.y > 0 ? 1 : -1;
linear_inputs[i * 2] = trainSpheresTransforms[i].position.x;
linear_inputs[(i * 2) + 1] = trainSpheresTransforms[i].position.z;
}
var matrix_inputs = RBFEntriesD(0.5, Y);
// Create Model
model = VisualStudioLibWrapper.linear_model_create(trainSpheresTransforms.Length);
// Train Model
VisualStudioLibWrapper.linear_model_train_classification(model, matrix_inputs, trainSpheresTransforms.Length, 2, Y, Y.Length, 1000000, 0.01f);
//double* model = {-0.4, 0.4, 0.2};
//double[] model = new double [] {-0.4, 0.4, 0.2};
// For each testSphere : Predict
foreach (var testSpheres in testSpheresTransforms)
{
double[] input = { testSpheres.position.x, testSpheres.position.z };
//double y = (float)(-model[1] / model[2] * testSpheresTransform.position.x - model[0] / model[2]);
float y = (float)VisualStudioLibWrapper.linear_model_predict_classification(model, input, 2);
testSpheres.position = new Vector3(
testSpheres.position.x,
y,
testSpheres.position.z
);
}
// Delete Model
//VisualStudioLibWrapper.clearArray(model);
}
public unsafe void RBFEntries()
{
double epsilon = 0.8;
int count = 0;
int blue_count = 0;
int red_count = 0;
// Créer dataset_inputs
// Créer dataest_expected_outputs
var Y = new double[trainSpheresTransforms.Length];
IntPtr model;
var linear_inputs = new double[trainSpheresTransforms.Length * 2];
for (int i = 0; i < trainSpheresTransforms.Length; i++)
{
Y[i] = trainSpheresTransforms[i].position.y > 0 ? 1 : -1;
linear_inputs[i * 2] = trainSpheresTransforms[i].position.x;
linear_inputs[(i * 2) + 1] = trainSpheresTransforms[i].position.z;
}
var Ox = new double[trainSpheresTransforms.Length, trainSpheresTransforms.Length];
//var Oz = new double[trainSpheresTransforms.Length,trainSpheresTransforms.Length];
var multiY = new double[Y.Length, 1];
for (int i = 0; i < trainSpheresTransforms.Length; i++)
{
for (int j = 0; j < trainSpheresTransforms.Length; j++)
{
Ox[i, j] = getRBFValue(epsilon, trainSpheresTransforms[i].position, trainSpheresTransforms[j].position);
//Oz[i,j] = getRBFValue(epsilon, trainSpheresTransforms[i].position.z, trainSpheresTransforms[j].position.z);
}
multiY[i, 0] = Y[i];
}
var l = trainSpheresTransforms.Length;
Matrix <double> OxMatrix = DenseMatrix.OfArray(Ox);
for (int i = 0; i < l; i++)
{
string str = "";
for (int j = 0; j < l; j++)
{
// Debug.Log(i + " " + j + ": " + Ox[i, j]);
str += OxMatrix[i, j] + "/ ";
}
Debug.Log(str);
}
//Matrix<double> OzMatrix = DenseMatrix.OfArray(Oz);
//Matrix<double> YMatrix = DenseMatrix.OfArray(Y);
// Debug.Log(OxMatrix.RowCount);
// Debug.Log(OxMatrix.ColumnCount);
Matrix <double> OxInv = OxMatrix.Inverse();
Matrix <double> mY = DenseMatrix.OfArray(multiY);
Matrix <double> resMatrix = OxInv.Multiply(mY);
Debug.Log("row = " + resMatrix.RowCount);
Debug.Log("col = " + resMatrix.ColumnCount);
var inputs = new double[resMatrix.RowCount];
for (int i = 0; i < resMatrix.RowCount; i++)
{
inputs[i] = resMatrix[i, 0];
}
model = VisualStudioLibWrapper.linear_model_create(2);
// Train Model
VisualStudioLibWrapper.linear_model_train_classification(model, inputs, 3, 1, Y, 3, 100000, 0.01f);
double *mdl = (double *)model.ToPointer();
for (int i = 0; i < 3; i++)
{
Debug.Log("model = " + mdl[i]);
}
var c1 = new Color(1, 2, 3);
var c2 = new Color(3, 2, 1);
foreach (var testSpheres in testSpheresTransforms)
{
double[] input = { testSpheres.position.x, testSpheres.position.z };
//double y = (float)(-model[1] / model[2] * testSpheresTransform.position.x - model[0] / model[2]);
float y = (float)VisualStudioLibWrapper.linear_model_predict_classification(model, input, 2);
// testSpheres.position = new Vector3(
// testSpheres.position.x,
// y,
// testSpheres.position.z
// );*
testSpheres.GetComponent <MeshRenderer>().material.color = y >= 0 ? c1 : c2;
}
// IntPtr model = VisualStudioLibWrapper.linear_model_create(2);
// VisualStudioLibWrapper.linear_model_train_classification(model, inputs, 3, 1, Y, 3, 100000, 0.01);
}
