func updateUntilOptimalSolutionsFound(lambda float64, m markov.MarkovChain, node *coupling.Node, exact [][]bool, visited [][]bool, d [][]float64, c coupling.Coupling, TPSolver func(markov.MarkovChain, *coupling.Node, [][]float64, float64, int, int), solvedNodes []*coupling.Node) {
log.Printf("find optimal for: (%v,%v)", node.S, node.T)
min, i, j := uvmethod.Run(node, d)
// if min is negative, we can further improve it, so we update it using the TPSolver and iterated until we cannot improve it further
for min < 0 && !utils.ApproxEqual(min, 0) {
previ, prevj := i, j
TPSolver(m, node, d, min, i, j)
setpair.Setpair(m, node, exact, visited, d, &c)
disc.Disc(lambda, node, exact, d, &c)
min, i, j = uvmethod.Run(node, d)
if previ == i && prevj == j && min < 0 {
break
}
}
// append solved nodes such that we do not end up recurively calling nodes that have already been found to be optimal
solvedNodes = append(solvedNodes, node)
for _, row := range node.Adj {
for _, edge := range row {
if edge.To.Adj == nil {
// if the node do not have an adjacency matrix or is exact, we do not have to proccess it
continue
}
if coupling.IsNodeInSlice(edge.To, solvedNodes) {
// if the node has already been proccesses, we do not have to do it again
continue
}
updateUntilOptimalSolutionsFound(lambda, m, edge.To, exact, visited, d, c, TPSolver, solvedNodes)
}
}
exact[node.S][node.T] = true
exact[node.S][node.T] = true
return
}
func TestOptimalSolutionFound(t *testing.T) {
c, m, visited, exact, d := coupling.SetUpTest()
d = sets.InitD(len(m.Transitions))
w := matching.FindFeasibleMatching(m, 0, 3, &c)
setpair.Setpair(m, w, exact, visited, d, &c)
// the expected solution is not precice since the markov chain used do not use precise fractions
expected := [][]float64{
[]float64{0, 0.33, 0},
[]float64{0, 0, 0.33},
[]float64{0.17, 0, 0},
[]float64{0.16, 0, 0.01}}
min, i, j := uvmethod.Run(w, d)
Solve(m, w, d, min, i, j)
for i := range w.Adj {
for j := range w.Adj[0] {
assert.True(t, utils.ApproxEqual(expected[i][j], w.Adj[i][j].Prob), "the optimal probability found was not what we expected")
}
}
}