public ClustersModel GetKmeanClusters(VrpData data, int clustersCount)
{
var centers = new Point2DReal[clustersCount];
var perm = Permutations.GetRandomPermutation(data.N - 1);
for (int i = 0; i < clustersCount; i++)
{
centers[i] = data.Customers[perm[i] + 1].Point;
}
var colors = new int[data.N];
var count = new int[clustersCount];
var changed = true;
while (changed)
{
changed = false;
count = new int[clustersCount];
for (int i = 1; i < data.N; i++)
{
var bestCenter = -1;
var bestDist = 0D;
for (int j = 0; j < clustersCount; j++)
{
if (bestCenter == -1 || centers[j].Dist(data.Customers[i].Point) < bestDist)
{
bestCenter = j;
bestDist = centers[j].Dist(data.Customers[i].Point);
}
}
if (colors[i] != bestCenter)
{
colors[i] = bestCenter;
changed = true;
}
}
for (int j = 0; j < clustersCount; j++)
{
var center = new Point2DReal(0, 0);
for (int i = 1; i < data.N; i++)
{
if (colors[i] == j)
{
center = center + data.Customers[i].Point;
count[j]++;
}
}
centers[j] = center * (1D / count[j]);
}
}
var result = new ClustersModel();
result.Color = colors;
result.Count = count;
result.Centers = centers;
return result;
}
private bool UpdateClustersWarehouse(VrpData data, ClustersModel clusters, int v)
{
var solver = new GreedyHeuristic();
var warehouseData = new WarehouseInputData();
warehouseData.N = v;
warehouseData.M = data.N;
warehouseData.T = new double[data.N,v];
for (int i = 0; i < data.N; i++)
{
for (int j = 0; j < v; j++)
{
warehouseData.T[i, j] = clusters.Centers[j].Dist(data.Customers[i].Point);
}
}
warehouseData.Consumers = data.Customers.Select(c => new Consumer {Demand = c.Demand, Id = c.Id}).ToArray();
warehouseData.Warehouses = new global::Warehouse.Warehouse[v];
for (int i = 0; i < v; i++)
{
warehouseData.Warehouses[i] = new global::Warehouse.Warehouse {Cap = data.C, Id = i, S = 0};
}
var result = solver.Solve(warehouseData);
if (result.SolutionFound)
{
clusters.Color = result.Solution;
return true;
}
return false;
}
private bool UpdateClustersMIP(VrpData data, ClustersModel baseClusters, int v)
{
throw new NotImplementedException();
}
// burn
private static bool UpdateClustersBurn(VrpData data, ClustersModel baseClusters, int v)
{
var rnd = new Random();
var found = false;
for (int count = 0; count < 10; count++)
{
var clusters = new ClustersModel();
clusters.Color = new int[data.N];
Array.Copy(baseClusters.Color, clusters.Color, data.N);
clusters.Count = new int[v];
Array.Copy(baseClusters.Count, clusters.Count, v);
clusters.Centers = new Point2DReal[v];
for (int i = 0; i < v; i++)
{
clusters.Centers[i] = new Point2DReal(baseClusters.Centers[i]);
}
var w = new int[v];
for (int i = 0; i < data.N; i++)
{
w[clusters.Color[i]] += data.Customers[i].Demand;
}
var baseOverweight = w.Sum(x => x > data.C ? x - data.C : 0);
var overweight = baseOverweight;
var radiuses = new double[v];
for (int i = 0; i < data.N; i++)
{
var color = clusters.Color[i];
var dist = data.Customers[i].Point.Dist(clusters.Centers[color]);
if (dist > radiuses[color])
{
radiuses[color] = dist;
}
}
var baseRadius = radiuses.Sum();
var radius = baseRadius;
var E = 1D;
var minE = double.MaxValue;
var bestColors = new int[data.N];
var bestW = new int[v];
for (int k = 0; k < 1000; k++)
{
if (E < minE)
{
minE = E;
Array.Copy(clusters.Color, bestColors, data.N);
Array.Copy(w, bestW, v);
}
var id = rnd.Next(data.N);
var weight = data.Customers[id].Demand;
var newColor = rnd.Next(v - 1);
var oldColor = clusters.Color[id];
if (oldColor <= newColor)
{
newColor++;
}
var weightDelta = -(w[newColor] + weight > data.C ? w[newColor] + weight - data.C : 0)
+ (w[newColor] > data.C ? w[newColor] - data.C : 0)
- (w[oldColor] - weight > data.C ? w[oldColor] - weight - data.C : 0)
+ (w[oldColor] > data.C ? w[oldColor] - data.C : 0);
var newOverweight = overweight - weightDelta;
var point = data.Customers[id].Point;
var newCenter = (clusters.Centers[newColor] * clusters.Count[newColor] + point)
* (1D / (clusters.Count[newColor] + 1));
var newDist = point.Dist(newCenter);
var newRadius = radius;
if (newDist > radiuses[newColor])
{
newRadius += newDist - radiuses[newColor];
}
var newE = (double) newOverweight / baseOverweight + K * (newRadius - baseRadius) / baseRadius;
var alpha = rnd.NextDouble();
var T = 10 / (k + 1);
var h = Math.Exp(-(newE - E) / T);
if (alpha < h)
{
E = newE;
overweight = newOverweight;
radius = newRadius;
w[newColor] += weight;
w[oldColor] -= weight;
clusters.Color[id] = newColor;
clusters.Centers[newColor] = newCenter;
clusters.Centers[oldColor] = (clusters.Centers[oldColor] * clusters.Count[oldColor] - point)
* (1D / (clusters.Count[oldColor] - 1));
clusters.Count[newColor]++;
clusters.Count[oldColor]--;
if (newDist > radiuses[newColor])
{
radiuses[newColor] = newDist;
}
}
}
if (bestW.All(x => x <= data.C))
{
found = true;
baseClusters.Color = bestColors;
break;
}
}
return found;
}
