public Classifiers.IClassifier learn_random_forest_on_known_points(Func <double[], double> meFunc, Func <double[], double[]> calcDerivative, double allowErr)
{
int[] count = new int[N]; for (int i = 0; i < N; i++)
{
count[i] = (Min[i] == Max[i]) ? 1 : NGRID;
}
create_grid(count);
analyse_voronoi();
analyse_error();
int n = xf.Length;
Classifiers.LabeledData[] ldata = new Classifiers.LabeledData[n];
int featureCount = 0;
for (int i = 0; i < n; i++)
{
double[] feature = build_fetures_from_existing_points(i, calcDerivative);
ldata[i] = new Classifiers.LabeledData(feature, 1);
featureCount = feature.Length;
}
Classifiers.IClassifier cls = new Classifiers.RandomForest();
Classifiers.RandomForestParams ps = new Classifiers.RandomForestParams(ldata, n /* samples count */,
featureCount /* features count */,
2 /* classes count */,
n / 10 /* trees count */,
5 /* count of features to do split in a tree */,
0.7 /* percent of a training set of samples */
/* used to build individual trees. */);
cls.train(ps);
double trainModelPrecision;
cls.validate(ldata, out trainModelPrecision);
Console.WriteLine("Model precision on training dataset: " + trainModelPrecision);
return(cls);
}
public void do_random_forest_analyse(Classifiers.IClassifier cls, double allowErr, Func <double[], double> meFunc, Func <double[], double[]> calcDerivative)
{
int[] count = new int[N]; for (int i = 0; i < N; i++)
{
count[i] = (Min[i] == Max[i]) ? 1 : NGRID;
}
create_grid(count);
analyse_voronoi();
analyse_error();
int n = candidates.Length;
Console.WriteLine(candidates.Length);
//int n = grid.Node.Length;
Classifiers.LabeledData[] ldata = new Classifiers.LabeledData[n];
int featureCount = 0;
for (int i = 0; i < n; i++)
{
// min, max in locality
double maxNeighbours = double.MinValue;
double minNeighbours = double.MaxValue;
foreach (var neighbour in grid.Neighbours(candidates[i]))
//foreach (var neighbour in grid.Neighbours(i))
{
double[] calcNeighbour = (double[])grid.Node[neighbour].Clone();
this.func.Calculate(calcNeighbour);
if (calcNeighbour[calcNeighbour.Length - 1] < minNeighbours)
{
minNeighbours = calcNeighbour[calcNeighbour.Length - 1];
}
if (calcNeighbour[calcNeighbour.Length - 1] > maxNeighbours)
{
maxNeighbours = calcNeighbour[calcNeighbour.Length - 1];
}
}
// current val
double[] cuurentNode = (double[])grid.Node[candidates[i]].Clone();
// double[] cuurentNode = (double[])grid.Node[i].Clone();
this.func.Calculate(cuurentNode);
double cuurentNodeVal = cuurentNode[cuurentNode.Length - 1];
if (cuurentNodeVal < minNeighbours)
{
minNeighbours = cuurentNodeVal;
}
if (cuurentNodeVal > maxNeighbours)
{
maxNeighbours = cuurentNodeVal;
}
// is real function and approximation are equal, class for point
//derivative
double[] derivative = calcDerivative(grid.Node[candidates[i]]);
//double[] derivative = calcDerivative(grid.Node[i]);
// build features vector
double[] features = new double[5 + derivative.Length];
features[0] = borderdist[i];
features[1] = error[i];
features[2] = maxNeighbours;
features[3] = minNeighbours;
features[4] = cuurentNodeVal;
for (int k = 0; k < derivative.Length; k++)
{
features[5 + k] = derivative[k];
}
ldata[i] = new Classifiers.LabeledData(features, 0);
featureCount = features.Length;
}
List <int> newCandidates = new List <int>();
int[] y = new int[ldata.Length];
for (int i = 0; i < ldata.Length; i++)
{
cls.infer(ldata[i].data, out y[i]);
if (y[i] == 1)
{
//newCandidates.Add(candidates[i]);
newCandidates.Add(i);
}
}
candidates = newCandidates.ToArray();
Console.WriteLine(candidates.Length);
xfcandidates = Tools.Sub(grid.Node, candidates);
}
public Classifiers.IClassifier learn_random_forest_on_grid(Func <double[], double> meFunc, Func <double[], double[]> calcDerivative, double allowErr)
{
int[] count = new int[N]; for (int i = 0; i < N; i++)
{
count[i] = (Min[i] == Max[i]) ? 1 : NGRID;
}
create_grid(count);
analyse_voronoi();
analyse_error();
int n = grid.Node.Length + xf.Length;
// int n = grid.Node.Length;
Classifiers.LabeledData[] ldata = new Classifiers.LabeledData[n];
int featureCount = 0;
for (int i = 0; i < grid.Node.Length; i++)
{
// min, max in locality
double maxNeighbours = double.MinValue;
double minNeighbours = double.MaxValue;
foreach (var neighbour in grid.Neighbours(i))
{
double[] calcNeighbour = (double[])grid.Node[neighbour].Clone();
this.func.Calculate(calcNeighbour);
if (calcNeighbour[calcNeighbour.Length - 1] < minNeighbours)
{
minNeighbours = calcNeighbour[calcNeighbour.Length - 1];
}
if (calcNeighbour[calcNeighbour.Length - 1] > maxNeighbours)
{
maxNeighbours = calcNeighbour[calcNeighbour.Length - 1];
}
}
// current val
double[] cuurentNode = (double[])grid.Node[i].Clone();
this.func.Calculate(cuurentNode);
double cuurentNodeVal = cuurentNode[cuurentNode.Length - 1];
if (cuurentNodeVal < minNeighbours)
{
minNeighbours = cuurentNodeVal;
}
if (cuurentNodeVal > maxNeighbours)
{
maxNeighbours = cuurentNodeVal;
}
// is real function and approximation are equal, class for point
int pointClass = 0;
if (Math.Abs(meFunc(grid.Node[i]) - cuurentNodeVal) > allowErr)
{
pointClass = 1;
}
//derivative
double[] derivative = calcDerivative(grid.Node[i]);
// build features vector
double[] features = new double[5 + derivative.Length];
features[0] = borderdist[i];
features[1] = error[i];
features[2] = maxNeighbours;
features[3] = minNeighbours;
features[4] = cuurentNodeVal;
for (int k = 0; k < derivative.Length; k++)
{
features[5 + k] = derivative[k];
}
ldata[i] = new Classifiers.LabeledData(features, pointClass);
featureCount = features.Length;
}
for (int i = 0; i < xf.Length; i++)
{
double[] feature = build_fetures_from_existing_points(i, calcDerivative);
ldata[grid.Node.Length + i] = new Classifiers.LabeledData(feature, 0);
featureCount = feature.Length;
}
Classifiers.IClassifier cls = new Classifiers.RandomForest();
Classifiers.RandomForestParams ps = new Classifiers.RandomForestParams(ldata, n /* samples count */,
featureCount /* features count */,
2 /* classes count */,
n / 10 /* trees count */,
6 /* count of features to do split in a tree */,
0.7 /* percent of a training set of samples */
/* used to build individual trees. */);
cls.train(ps);
double trainModelPrecision;
cls.validate(ldata, out trainModelPrecision);
Console.WriteLine("Model precision on training dataset: " + trainModelPrecision);
return(cls);
}
private void analyse_voronoi()
{
//вычисляю принадлежность узлов сетки доменам (графовый алгоритм на базе структуры уровней смежности)
SortedSet <int>[] adjncy = new SortedSet <int> [xf.Length];
for (int i = 0; i < xf.Length; i++)
{
adjncy[i] = new SortedSet <int>();
}
Console.WriteLine("Разбиение пространства на домены");
Queue <int> queue = new Queue <int>();
domain = new int[grid.Node.Length];
double[] dist = new double[grid.Node.Length];
for (int i = 0; i < domain.Length; i++)
{
domain[i] = -1;
dist[i] = double.PositiveInfinity;
}
for (int i = 0; i < xf.Length; i++)
{
int index;
grid.ToIndex(xf[i], out index);
dist[index] = distanceX(grid.Node[index], xf[i]);
domain[index] = i;
//setvalue(index, xf[i]);
queue.Enqueue(index);
}
while (queue.Count > 0)
{
int index = queue.Dequeue();
int i = domain[index];
foreach (var adj in grid.Neighbours(index))
{
double d = distanceX(grid.Node[adj], xf[i]);
if (domain[adj] >= 0)
{
adjncy[domain[adj]].Add(i);
adjncy[i].Add(domain[adj]);
if (d < dist[adj])
{
domain[adj] = i;
dist[adj] = d;
}
continue;
}
domain[adj] = i;
dist[adj] = d;
//setvalue(adj, xf[i]);
queue.Enqueue(adj);
}
}
Console.WriteLine("Построение графа доменов");
//строю граф соседства доменов
graph = new int[xf.Length][];
for (int i = 0; i < xf.Length; i++)
{
adjncy[i].Add(i);
graph[i] = adjncy[i].ToArray();
}
Console.WriteLine("Построение диграммы Вороного на сетке");
//уточняю домены (диаграмма вороного на сетке)
for (int i = 0; i < grid.Node.Length; i++)
{
double[] xy = grid.Node[i];
int[] adj = graph[domain[i]];
double min = double.PositiveInfinity;
for (int j = 0; j < adj.Length; j++)
{
double d = distanceX(xy, xf[adj[j]]);
if (min > d)
{
min = d; domain[i] = adj[j];
}
}
}
Console.WriteLine("Вычисление границ доменов");
//вычисляю границы доменов
borderdist = new double[grid.Node.Length];
bordernear = new int[grid.Node.Length];
for (int i = 0; i < grid.Node.Length; i++)
{
borderdist[i] = double.PositiveInfinity;
bordernear[i] = -1;
int dom = domain[i];
foreach (var adj in grid.Neighbours(i))
{
if (domain[adj] != dom)
{
borderdist[i] = 0;
bordernear[i] = i;
queue.Enqueue(i);
break;
}
}
}
candidates = queue.ToArray();
//--------------------------TO REMOVE AFTER PROPER CLASSIFIER USAGE-------------------------
//--------------------------------------Too silly example-----------------------------------
// Classification is binary.. Shall we use more simplest binary classifier?
Classifiers.LabeledData[] ldata = new Classifiers.LabeledData[3];
ldata[0] = new Classifiers.LabeledData(new double[3] {
grid.Node[0][0], borderdist[0], bordernear[0]
}, 1);
ldata[1] = new Classifiers.LabeledData(new double[3] {
grid.Node[1][0], borderdist[1], bordernear[1]
}, 1);
ldata[2] = new Classifiers.LabeledData(new double[3] {
grid.Node[2][0], borderdist[2], bordernear[2]
}, 0);
Classifiers.IClassifier cls = new Classifiers.RandomForest();
Classifiers.RandomForestParams ps = new Classifiers.RandomForestParams(ldata, 3 /* samples count */,
3 /* features count */,
2 /* classes count */,
3 /* trees count */,
2 /* count of features to do split in a tree */,
0.7 /* percent of a training set of samples */
/* used to build individual trees. */);
cls.train(ps);
int[] y = new int[3];
cls.infer(ldata[0].data, out y[0]);
cls.infer(ldata[1].data, out y[1]);
cls.infer(ldata[2].data, out y[2]);
for (int i = 0; i < 3; i++)
{
Console.WriteLine("{0} is predicted y[{1}] from trained data sample and {2} is ground truth", y[i], i, ldata[i].label);
}
double trainModelPrecision;
cls.validate(ldata, out trainModelPrecision);
Console.WriteLine("Model precision on training dataset: " + trainModelPrecision);
//------------------------------------------------------------------------------------------
//--------------------------TO REMOVE AFTER PROPER CLASSIFIER USAGE-------------------------
Console.WriteLine("Построение функции расстояний до границ доменов");
//вычисляю расстояния от границ
while (queue.Count > 0)
{
int index = queue.Dequeue();
int dom = domain[index];
int brd = bordernear[index];
foreach (var adj in grid.Neighbours(index))
{
if (domain[adj] != dom)
{
continue;
}
double d = distanceX(grid.Node[adj], grid.Node[brd]);
if (bordernear[adj] >= 0)
{
if (d < borderdist[adj])
{
bordernear[adj] = brd;
borderdist[adj] = d;
}
continue;
}
bordernear[adj] = brd;
borderdist[adj] = d;
queue.Enqueue(adj);
}
}
Console.WriteLine("Нормировка функции расстояний до границ доменов");
//нормирую расстояния от границ
for (int i = 0; i < grid.Node.Length; i++)
{
int dom = domain[i];
int brd = bordernear[i];
double a = distanceX(grid.Node[i], xf[dom]);
double b = distanceX(grid.Node[i], grid.Node[brd]);
double c = a + b;
borderdist[i] = (c == 0) ? 0 : b / c;
}
}
