public void ResolveClassificationMLP()
{
object myModel = MyFirstDLLWrapper.mlp_create_model(3, new int[] { 2, 3, 1 });
double[][] inputs = new double[blueSamples.Count + redSamples.Count][];
double[] outputs = new double[blueSamples.Count + redSamples.Count];
int i = 0;
foreach (var blue in blueSamples)
{
inputs[i] = new double[2];
inputs[i][0] = blue.transform.position.x / 10;
inputs[i][1] = blue.transform.position.z / 10;
outputs[i++] = 1;
}
foreach (var red in redSamples)
{
inputs[i] = new double[2];
inputs[i][0] = red.transform.position.x / 10;
inputs[i][1] = red.transform.position.z / 10;
outputs[i++] = -1;
}
MyFirstDLLWrapper.mlp_fit_classification_backdrop(myModel, inputs, outputs, 100000, 0.01);
foreach (var white in whiteSamples)
{
Destroy(white);
}
whiteSamples.Clear();
for (int a = 1; a <= 10; a++)
{
for (int b = 1; b <= 10; b++)
{
var rslt = MyFirstDLLWrapper.mlp_classify(myModel, new double[] { a / 10f, b / 10f });
if (rslt[0] > 0)
{
whiteSamples.Add(Instantiate(bluePrefab, new Vector3(a, 0, b), Quaternion.identity));
}
else
{
whiteSamples.Add(Instantiate(redPrefab, new Vector3(a, 0, b), Quaternion.identity));
}
}
}
}
public void ResolveClassification()
{
double[,] inputs = new double[blueSamples.Count + redSamples.Count, 2];
double[] outputs = new double[blueSamples.Count + redSamples.Count];
int i = 0;
foreach (var blue in blueSamples)
{
inputs[i, 0] = blue.transform.position.x;
inputs[i, 1] = blue.transform.position.z;
outputs[i++] = 1;
}
foreach (var red in redSamples)
{
inputs[i, 0] = red.transform.position.x;
inputs[i, 1] = red.transform.position.z;
outputs[i++] = -1;
}
MyFirstDLLWrapper.linear_fit_classification_rosenblatt(ref myModelClassification, inputs, outputs, 1000, 0.1);
foreach (var white in whiteSamples)
{
Destroy(white);
}
whiteSamples.Clear();
for (int a = 1; a <= 10; a++)
{
for (int b = 1; b <= 10; b++)
{
var rslt = MyFirstDLLWrapper.linear_classify(ref myModelClassification, new double[] { a, b });
if (rslt > 0)
{
whiteSamples.Add(Instantiate(bluePrefab, new Vector3(a, 0, b), Quaternion.identity));
}
else
{
whiteSamples.Add(Instantiate(redPrefab, new Vector3(a, 0, b), Quaternion.identity));
}
}
}
}
public void GenerateCross()
{
foreach (var blue in blueSamples)
{
Destroy(blue);
}
foreach (var red in redSamples)
{
Destroy(red);
}
foreach (var plotPoint in plotSamples)
{
Destroy(plotPoint);
}
redSamples.Clear();
blueSamples.Clear();
plotSamples.Clear();
for (var i = 0; i < trainingSetLength; i++)
{
var xRdm = Random.Range(-10f, 10f);
var zRdm = Random.Range(-10f, 10f);
if (Mathf.Abs(xRdm) <= 2.5f || Mathf.Abs(zRdm) <= 2.5f)
{
blueSamples.Add(
Instantiate(bluePrefab, new Vector3(xRdm, 0f, zRdm), Quaternion.identity));
}
else
{
redSamples.Add(
Instantiate(redPrefab, new Vector3(xRdm, 0f, zRdm), Quaternion.identity));
}
}
for (var x = -10; x <= 10; x++)
{
for (var z = -10; z <= 10; z++)
{
plotSamples.Add(
Instantiate(plotPrefab, new Vector3(x, -2f, z), Quaternion.identity));
}
}
myModelClassification = MyFirstDLLWrapper.linear_create_model(2);
}
public void ResolveRegressionMLP()
{
object myModel = MyFirstDLLWrapper.mlp_create_model(3, new int[] { 2, 20, 1 });
double[][] inputs = new double[blueSamples.Count + redSamples.Count][];
double[] outputs = new double[blueSamples.Count + redSamples.Count];
int i = 0;
foreach (var blue in blueSamples)
{
inputs[i] = new double[2];
inputs[i][0] = blue.transform.position.x / 10;
inputs[i][1] = blue.transform.position.z / 10;
outputs[i++] = 0;
}
foreach (var red in redSamples)
{
inputs[i] = new double[2];
inputs[i][0] = red.transform.position.x / 10;
inputs[i][1] = red.transform.position.z / 10;
outputs[i++] = 1;
}
MyFirstDLLWrapper.mlp_fit_regression_backdrop(myModel, inputs, outputs, 10000, 0.1);
foreach (var white in whiteSamples)
{
Destroy(white);
}
whiteSamples.Clear();
for (int a = 1; a <= 10; a++)
{
for (int b = 1; b <= 10; b++)
{
var rslt = MyFirstDLLWrapper.mlp_predict(myModel, new double[] { a / 10f, b / 10f });
var elt = Instantiate(whitePrefab, new Vector3(a, 0f, b), Quaternion.identity);
elt.GetComponent <Renderer>().material.color = new Color((float)rslt[0], 0, 1 - (float)rslt[0]);
whiteSamples.Add(elt);
}
}
}
public void ResolveLinear()
{
double[] myModel = MyFirstDLLWrapper.linear_create_model(2);
double[,] inputs = new double[blueSamples.Count + redSamples.Count, 2];
double[] outputs = new double[blueSamples.Count + redSamples.Count];
int i = 0;
foreach (var blue in blueSamples)
{
inputs[i, 0] = blue.transform.position.x;
inputs[i, 1] = blue.transform.position.z;
outputs[i++] = 0;
}
foreach (var red in redSamples)
{
inputs[i, 0] = red.transform.position.x;
inputs[i, 1] = red.transform.position.z;
outputs[i++] = 1;
}
MyFirstDLLWrapper.linear_fit_regression(ref myModel, inputs, outputs);
foreach (var white in whiteSamples)
{
Destroy(white);
}
whiteSamples.Clear();
for (int a = 1; a <= 10; a++)
{
for (int b = 1; b <= 10; b++)
{
var rslt = MyFirstDLLWrapper.linear_predict(ref myModel, new double[] { a, b });
var elt = Instantiate(whitePrefab, new Vector3(a, 0f, b), Quaternion.identity);
elt.GetComponent <Renderer>().material.color = new Color((float)rslt, 0, 1 - (float)rslt);
whiteSamples.Add(elt);
}
}
}
// Use this for initialization
void Start()
{
myModelClassification = MyFirstDLLWrapper.linear_create_model(2);
}
