private void TrainModel(System.IntPtr model, double[,] values)
{
var nbInputs = values.GetUpperBound(0) - 1;
for (int y = 0; y < values.GetUpperBound(1); y++)
{
double[] inputs = new double[nbInputs];
int expected = 0;
for (int x = 0; x < values.GetUpperBound(0); x++)
{
var value = values[x, y];
if (x == nbInputs)
{
expected = Convert.ToInt32(value);
}
else
{
inputs[x] = value;
}
}
PanebWrapper.classification_train(model, inputs.Length, inputs, expected);
}
}
private void ComputeInputs(System.IntPtr model, double[,] values)
{
var nbInputs = values.GetUpperBound(0) - 1;
for (int y = 0; y < values.GetUpperBound(1); y++)
{
double[] inputs = new double[nbInputs];
int expected = 0;
for (int x = 0; x < values.GetUpperBound(0); x++)
{
var value = values[x, y];
if (x == nbInputs)
{
expected = Convert.ToInt32(value);
}
else
{
inputs[x] = value;
}
}
int result = PanebWrapper.classification_compute(model, inputs.Length, inputs);
Debug.Log("Expected: " + expected + "; Result = " + result);
}
}
void Start()
{
var model = PanebWrapper.pmc_create(nbLayers, layers);
TrainModel(model);
MoveAxis(model);
}
void Start()
{
var weights = PanebWrapper.classification_create(3);
TrainModel(weights);
MoveAxis(weights);
}
private void ComputeInputs(System.IntPtr model, double[,] values)
{
var nbInputs = values.GetUpperBound(0) - 1;
for (int y = 0; y < values.GetUpperBound(1); y++)
{
double[] inputs = new double[nbInputs];
double expected = 0.0;
for (int x = 0; x < values.GetUpperBound(0); x++)
{
var value = values[x, y];
if (x == nbInputs)
{
expected = value;
}
else
{
inputs[x] = value;
}
}
double result = PanebWrapper.regression_point(model, inputs.Length, inputs);
Debug.Log("Expected: " + expected + "; Result = " + result);
}
}
private System.IntPtr CreateModel(double[,] values)
{
var nbInputs = values.GetUpperBound(0) - 1;
var nbInputRows = values.GetUpperBound(1);
var nbInputCols = values.GetUpperBound(0);
double[] inputs = new double[nbInputRows * nbInputCols];
var inputIterator = 0;
var nbOutputRows = values.GetUpperBound(1);
var nbOutputCols = 1;
double[] outputs = new double[nbOutputRows * nbOutputCols];
var outputIterator = 0;
for (int y = 0; y < values.GetUpperBound(1); y++)
{
for (int x = 0; x < values.GetUpperBound(0); x++)
{
var value = values[x, y];
if (x == nbInputs)
{
outputs[outputIterator++] = value;
}
else
{
inputs[inputIterator++] = value;
}
}
}
return(PanebWrapper.regression_compute(nbInputRows, nbInputCols, inputs, nbOutputRows, nbOutputRows, outputs));
}
private void PrintWeights(System.IntPtr weights)
{
Debug.Log(
PanebWrapper.classification_weights(weights, 0) + "; " +
PanebWrapper.classification_weights(weights, 1) + "; " +
PanebWrapper.classification_weights(weights, 2)
);
}
void Start()
{
var inputs = ComputeInputs();
var outputs = ComputeOutputs();
var weights = PanebWrapper.regression_compute(spheres.Length, 3, inputs, spheres.Length, 1, outputs);
MoveAxis(weights);
}
private void MoveAxis(System.IntPtr weights)
{
foreach (var white in whites)
{
var x = white.position.x;
var z = white.position.z;
int direction = PanebWrapper.classification_compute(weights, 2, new double[] { x, z });
white.position += direction == 1 ? Vector3.up : Vector3.down;
}
}
private void MoveAxis(System.IntPtr weights)
{
foreach (var white in whites)
{
var x = white.position.x;
var z = white.position.z;
double result = PanebWrapper.regression_point(weights, 2, new double[] { x, z });
//Debug.Log("[" + x + ", " + z + "] = " + result);
white.position += Vector3.up * (float)result;
}
}
private void MoveAxis(System.IntPtr model)
{
foreach (var white in whites)
{
var x = white.position.x;
var z = white.position.z;
System.IntPtr outputs = PanebWrapper.pmc_compute(nbLayers, layers, model, 2, new double[] { x, z }, 0);
double value = PanebWrapper.pmc_value(outputs, 0);
white.position += value > 0.0 ? Vector3.up : Vector3.down;
}
}
void Start()
{
string[,] csvTraining = CSVReader.SplitCsvGrid(trainingCsv.text);
double[,] trainingValues = StringArrayToDouble(csvTraining);
string[,] csvTest = CSVReader.SplitCsvGrid(testCsv.text);
double[,] testValues = StringArrayToDouble(csvTest);
var nbWeights = trainingValues.GetUpperBound(0);
var model = PanebWrapper.classification_create(nbWeights);
TrainModel(model, trainingValues);
ComputeInputs(model, testValues);
}
private void TrainModel(System.IntPtr weights)
{
for (int i = 0; i < 300; ++i)
{
foreach (var sphere in spheres)
{
var transform = sphere.GetComponent <Transform>();
var x = transform.position.x;
var z = transform.position.z;
var expected = (int)transform.position.y;
PanebWrapper.classification_train(weights, 2, new double[] { x, z }, expected);
}
}
}
private void TrainModel(System.IntPtr model)
{
for (int i = 0; i < 300; ++i)
{
foreach (var sphere in spheres)
{
var transform = sphere.GetComponent <Transform>();
var x = transform.position.x;
var z = transform.position.z;
var expected = transform.position.y;
PanebWrapper.pmc_train(nbLayers, layers, model, 2, new double[] { x, z }, 1, new double[] { expected }, 0);
}
}
}