00001
00002
00003 class dir_cut : public cons_obj {
00004 private:
00005 var_map<edge>& VMy;
00006 var_map<edge>& VMz;
00007 graph& G;
00008 edge_array<double>& D;
00009 list<row> NZ;
00010
00011 public:
00012
00013 dir_cut(graph& G_, node_array<double>& C, var_map<edge>& VMy_,
00014 var_map<edge>& VMz_, edge_array<double>& D_, GraphWin& GW)
00015 : cons_obj(Greater, 1), VMy(VMy_), VMz(VMz_), G(G_), D(D_) {
00016 node_array<double> CD(G);
00017 node u;
00018 forall_nodes(u, G) CD[u]=C[u];
00019 edge e;
00020 forall_edges(e, G) {
00021 if((CD[source(e)]>0) && (CD[target(e)]==0) && (CD[source(e)]>D[e]))
00022 CD[source(e)]=D[e];
00023 if((CD[target(e)]>0) && (CD[source(e)]==0) && (CD[target(e)]>D[e]))
00024 CD[target(e)]=D[e];
00025 }
00026 forall_edges(e, G) {
00027
00028 if(coeff_y(e, CD)>0) {
00029 NZ.append(VMy[e]);
00030
00031 }
00032 if(coeff_z(e, CD)>0) {
00033 NZ.append(VMz[e]);
00034
00035 }
00036 };
00037
00038 };
00039
00040 double coeff_y(edge e, node_array<double>& CD) {
00041 if((CD[source(e)]>0) && (CD[source(e)]<=D[e])) return 1;
00042 return 0;
00043 };
00044
00045 double coeff_z(edge e, node_array<double>& CD) {
00046 if((CD[target(e)]>0) && (CD[target(e)]<=D[e])) return 1;
00047 return 0;
00048 };
00049
00050 virtual void non_zero_entries(row& r) {
00051 row v;
00052 forall(v, NZ) r+=v;
00053 };
00054 };
00055
00056 class Broadcast : public sym_constraint {
00057
00058 private:
00059 graph& G;
00060 graph GD;
00061 node_array<node> GtoGD;
00062 edge_array<edge> REV;
00063 var_map<edge>& VM;
00064 var_map<edge>& VMy;
00065 var_map<edge>& VMz;
00066 var_map<edge> VMD;
00067 GraphWin& GW;
00068 edge_array<double>& C;
00069
00070 public:
00071
00072 Broadcast(graph& G_, var_map<edge>& VM_, var_map<edge>& VMy_,
00073 var_map<edge>& VMz_, GraphWin& GW_, edge_array<double>& C_)
00074 : G(G_), VM(VM_), VMy(VMy_), VMz(VMz_), GW(GW_), C(C_) {
00075 };
00076
00077 void init(subproblem& s) {
00078 node_array<double> CD(G, 0);
00079
00080 GtoGD.init(G);
00081 node u;
00082 forall_nodes(u, G)
00083 GtoGD[u]=GD.new_node();
00084
00085 edge e,f,g;
00086 edge_array<double> C1(GD, 2*G.number_of_edges(), 0);
00087 REV.init(GD, 2*G.number_of_edges(), e);
00088 forall_edges(e, G) {
00089 f=GD.new_edge(GtoGD[source(e)], GtoGD[target(e)]);
00090 VMD[f]=VMy[e];
00091 C1[f]=-C[e];
00092 g=GD.new_edge(GtoGD[target(e)], GtoGD[source(e)]);
00093 VMD[g]=VMz[e];
00094 C1[g]=-C[e];
00095 REV[g]=f;
00096 REV[f]=g;
00097 };
00098
00099 GD.sort_edges(C1);
00100
00101
00102
00103
00104
00105
00106
00107
00108
00109 };
00110
00111 void mc_dfs(node u, edge_array<int>& c, edge_array<int>& f, node_array<bool>& r) {
00112 r[u]=true;
00113 edge e;
00114 forall_out_edges(e, u)
00115 if((!r[target(e)]) && (f[e]!=c[e]))
00116 mc_dfs(target(e), c, f, r);
00117 };
00118
00119 status separate(subproblem& s) {
00120 edge_array<int> cap(GD);
00121 edge e;
00122
00123 status st=no_constraint_found;
00124
00125 node_array<double> SE(GD, 0);
00126
00127
00128 forall_edges(e, GD) {
00129 cap[e]=(int) (1000*(s.value(VMD[e])+SE[source(e)]));
00130 SE[source(e)]+=s.value(VMD[e]);
00131 };
00132
00133 edge_array<int> capr(GD);
00134 forall_edges(e, GD) capr[e]=cap[REV[e]];
00135
00136 node u, v, w;
00137 int k;
00138 edge_array<int> f(GD);
00139 u=GD.choose_node();
00140 forall_nodes(v, GD) if (u!=v) {
00141 {
00142 k=MAX_FLOW(GD, u, v, cap, f);
00143
00144 node_array<bool> r(GD, false);
00145 mc_dfs(u, cap, f, r);
00146 node_array<double> r1(G);
00147 forall_nodes(w, G) if(r[GtoGD[w]]) r1[w]=MAX_DOUBLE; else r1[w]=0;
00148 cons_obj* c=new dir_cut(G, r1, VMy, VMz, C, GW);
00149 if(c->violated(s)) {
00150
00151 s.add_basic_constraint(c);
00152 st=constraint_found;
00153 };
00154 }
00155 {
00156 k=MAX_FLOW(GD, u, v, capr, f);
00157
00158 node_array<bool> r(GD, false);
00159 mc_dfs(u, capr, f, r);
00160 node_array<double> r1(G);
00161 forall_nodes(w, G) if(!r[GtoGD[w]]) r1[w]=MAX_DOUBLE; else r1[w]=0;
00162 cons_obj* c=new dir_cut(G, r1, VMy, VMz, C, GW);
00163 if(c->violated(s)) {
00164
00165 s.add_basic_constraint(c);
00166 st=constraint_found;
00167 };
00168 }
00169 }
00170
00171 return st;
00172 }
00173
00174
00175
00176
00177
00178
00179
00180
00181
00182
00183
00184
00185
00186
00187
00188
00189
00190
00191
00192
00193
00194
00195
00196
00197
00198
00199
00200
00201
00202
00203
00204
00205
00206
00207
00208
00209
00210 };
00211
00212 double ComputeBroadcast(graph& G, edge_array<double>& Cost, list<edge>& T,
00213 GraphWin& GW) {
00214
00215 list<edge> MST=MIN_SPANNING_TREE(G, Cost);
00216 node_array<double> MSTC(G,0);
00217 edge e;
00218 forall(e,MST) {
00219 MSTC[source(e)]=leda_max(MSTC[source(e)], Cost[e]);
00220 MSTC[target(e)]=leda_max(MSTC[target(e)], Cost[e]);
00221 };
00222 node u; double mstc=0;
00223 forall_nodes(u, G) mstc+=MSTC[u]*MSTC[u];
00224
00225
00226
00227 ILP_Problem IP(Optsense_Min);
00228
00229 edge f; e=nil;
00230 var_map<edge> VM;
00231 var_map<edge> VMy;
00232 var_map<edge> VMz;
00233
00234 forall_edges(e,G) {
00235 VM[e]=IP.add_variable(-0.1, 0, 1, Vartype_Integer);
00236 VMy[e]=IP.add_variable(Cost[e]*Cost[e], 0, 1, Vartype_Integer);
00237 VMz[e]=IP.add_variable(Cost[e]*Cost[e], 0, 1, Vartype_Integer);
00238 }
00239
00240 forall_edges(e,G) {
00241 row r1;
00242 forall_out_edges(f,source(e)) if(Cost[f]>=Cost[e]-0.001) r1+=VMy[f];
00243 forall_in_edges(f, source(e)) if(Cost[f]>=Cost[e]-0.001) r1+=VMz[f];
00244 IP.add_basic_constraint(r1 >= VM[e]);
00245
00246 row r2;
00247 forall_out_edges(f,target(e)) if(Cost[f]>=Cost[e]-0.001) r2+=VMy[f];
00248 forall_in_edges(f, target(e)) if(Cost[f]>=Cost[e]-0.001) r2+=VMz[f];
00249 IP.add_basic_constraint(r2 >= VM[e]);
00250 }
00251
00252 forall_nodes(u, G) {
00253 row r;
00254 forall_out_edges(e, u) r+=VMy[e];
00255 forall_in_edges(e, u) r+=VMz[e];
00256 IP.add_basic_constraint(r==1);
00257 };
00258
00259 IP.add_sym_constraint(new SpanTree(G,VM));
00260
00261
00262
00263
00264
00265 IP.optimize();
00266
00267 double optc=0;
00268 forall_edges(e, G) {
00269 if(IP.get_solution(VM[e])>0.5) {
00270 T.append(e);
00271 };
00272 }
00273 forall_nodes(u, G) MSTC[u]=0;
00274 forall(e,T) {
00275 MSTC[source(e)]=leda_max(MSTC[source(e)], Cost[e]);
00276 MSTC[target(e)]=leda_max(MSTC[target(e)], Cost[e]);
00277 };
00278 forall_nodes(u, G) optc+=MSTC[u]*MSTC[u];
00279
00280 cout<<"Number of nodes "<<G.number_of_nodes()<<endl;
00281 cout<<"Cost of mst "<<mstc<<"\n";
00282 cout<<"Cost of optimal solution "<<optc<<"\n";
00283
00284 cout<<"Save "<<(mstc-optc)/mstc*100<<"\n";
00285
00286
00287 return (mstc-optc)/mstc*100;
00288 }