Skip to content
GitLab
Commit 0d0f4966 authored by Tiago Peixoto's avatar Tiago Peixoto
Browse files

* graph_community.cc: code parallelization and speed improvements 

* graph/graph_distance.cc: trivial "one-liner" speed improvement



git-svn-id: https://svn.forked.de/graph-tool/trunk@98 d4600afd-f417-0410-95de-beed9576f240
parent 54b33836
Hide whitespace changes
Inline Side-by-side
src/graph/graph_community.cc
View file @ 0d0f4966
......@@ -50,42 +50,6 @@ size_t out_degree_no_loops(typename graph_traits<Graph>::vertex_descriptor v, co
}
//==============================================================================
// ManagedUnorderedMap
//==============================================================================
template <class Key, class Value>
class ManagedUnorderedMap: public tr1::unordered_map<Key,Value>
{
typedef tr1::unordered_map<Key,Value> base_t;
public:
void erase(typename base_t::iterator pos)
{
static_cast<base_t*>(this)->erase(pos);
manage();
}
size_t erase(const Key& k)
{
size_t n = static_cast<base_t*>(this)->erase(k);
manage();
return n;
}
void manage()
{
if (this->bucket_count() > 2*this->size())
{
base_t* new_map = new base_t;
for (typeof(this->begin()) iter = this->begin(); iter != this->end(); ++iter)
(*new_map)[iter->first] = iter->second;
*static_cast<base_t*>(this) = *new_map;
delete new_map;
}
}
};
//==============================================================================
// GetCommunityStructure()
// computes the community structure through a spin glass system with
......@@ -98,7 +62,8 @@ struct get_communities
template <class Graph, class WeightMap, class CommunityMap>
void operator()(Graph& g, WeightMap weights, CommunityMap s, double gamma, size_t n_iter, pair<double,double> Tinterval, pair<size_t,bool> Nspins, size_t seed, pair<bool,string> verbose) const
{
size_t N = HardNumVertices()(g);
stringstream out_str;
ofstream out_file;
if (verbose.second != "")
......@@ -118,8 +83,8 @@ struct get_communities
if (Nspins.first == 0)
Nspins.first = HardNumVertices()(g);
ManagedUnorderedMap<size_t, size_t> Ns; // spin histogram
ManagedUnorderedMap<size_t, map<double, unordered_set<size_t> > > global_term; // global energy term
unordered_map<size_t, size_t> Ns; // spin histogram
unordered_map<size_t, map<double, unordered_set<size_t> > > global_term; // global energy term
// init spins from [0,N-1] and global info
uniform_int<size_t> sample_spin(0, Nspins.first-1);
......@@ -160,13 +125,23 @@ struct get_communities
// calculate the cumulative probabilities of each spin energy level
unordered_map<size_t, map<long double, pair<double, bool> > > cumm_prob;
unordered_map<size_t, unordered_map<double, long double> > energy_to_prob;
for (size_t i = 0; i < degs.size(); ++i)
int i, NK = degs.size();
#ifdef USING_OPENMP
for (i = 0; i < NK; ++i)
{
cumm_prob[degs[i]];
energy_to_prob[degs[i]];
}
#endif //USING_OPENMP
#pragma omp parallel for default(shared) private(i) schedule(dynamic)
for (i = 0; i < NK; ++i)
{
long double prob = 0;
for (typeof(global_term[degs[i]].begin()) iter = global_term[degs[i]].begin(); iter != global_term[degs[i]].end(); ++iter)
{
long double M = log(numeric_limits<long double>::max()/(Nspins.first*10));
long double this_prob = exp((long double)(-iter->first - degs[i])/T + M)*iter->second.size();
if (prob + this_prob != prob)
{
prob += this_prob;
......@@ -179,28 +154,39 @@ struct get_communities
}
}
if (prob == 0.0)
steepest_descent = true;
{
#pragma omp critical
{
steepest_descent = true;
}
}
}
// list of spins which were updated
vector<pair<typename graph_traits<Graph>::vertex_descriptor,size_t> > spin_update;
spin_update.reserve(HardNumVertices()(g));
spin_update.reserve(N);
// sample a new spin for every vertex
for (tie(v,v_end) = vertices(g); v != v_end; ++v)
int NV = num_vertices(g);
#pragma omp parallel for default(shared) private(i) firstprivate(global_term, cumm_prob) schedule(dynamic)
for (i = 0; i < NV; ++i)
{
typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);
if (v == graph_traits<Graph>::null_vertex())
continue;
unordered_map<size_t, double> ns; // number of neighbours with spin 's' (weighted)
// neighborhood spins info
typename graph_traits<Graph>::out_edge_iterator e,e_end;
for (tie(e,e_end) = out_edges(*v,g); e != e_end; ++e)
for (tie(e,e_end) = out_edges(v,g); e != e_end; ++e)
{
typename graph_traits<Graph>::vertex_descriptor t = target(*e,g);
if (t != *v)
if (t != v)
ns[s[t]] += get(weights, *e);
}
size_t k = out_degree_no_loops(*v,g);
size_t k = out_degree_no_loops(v,g);
map<double,unordered_set<size_t> >& global_term_k = global_term[k];
map<long double,pair<double,bool> >& cumm_prob_k = cumm_prob[k];
......@@ -228,7 +214,7 @@ struct get_communities
cumm_prob_k[energy_to_prob[k][*iter]].second = false;
if (global_term_k.find(*iter) != global_term_k.end())
{
long double M = log(numeric_limits<long double>::max()/(Nspins.first*10));
long double M = log(numeric_limits<long double>::max()/(Nspins.first*10));
long double prob = exp((long double)(-*iter - k)/T + M)*global_term_k[*iter].size();
if (cumm_prob_k.empty() || cumm_prob_k.rbegin()->first + prob != cumm_prob_k.rbegin()->first)
{
......@@ -256,7 +242,13 @@ struct get_communities
bool accept = false;
while (!accept)
{
typeof(cumm_prob_k.begin()) upper = cumm_prob_k.upper_bound(prob_sample(rng));
typeof(cumm_prob_k.begin()) upper;
#pragma omp critical
{
upper = cumm_prob_k.upper_bound(prob_sample(rng));
}
if (upper == cumm_prob_k.end())
upper--;
E = upper->second.first;
......@@ -267,7 +259,13 @@ struct get_communities
//new spin (randomly chosen amongst those with equal energy)
uniform_int<size_t> sample_spin(0,global_term_k[E].size()-1);
typeof(global_term_k[E].begin()) iter = global_term_k[E].begin();
advance(iter, sample_spin(rng));
size_t spin_n;
#pragma omp critical
{
spin_n = sample_spin(rng);
}
advance(iter, spin_n);
size_t a = *iter;
// cleanup modified probabilities
......@@ -288,16 +286,20 @@ struct get_communities
{
double new_E = gamma*Nnnks(k, iter->first);
double old_E = new_E - ns[iter->first];
global_term_k[old_E].erase(iter->first);
if (global_term_k[old_E].empty())
global_term_k.erase(old_E);
global_term_k[new_E].insert(iter->first);
}
//update global info
if (s[*v] != a)
spin_update.push_back(make_pair(*v, a));
if (s[v] != a)
{
#pragma omp critical
{
spin_update.push_back(make_pair(v, a));
}
}
}
// flip spins and update Nnnks
......@@ -306,30 +308,40 @@ struct get_communities
typename graph_traits<Graph>::vertex_descriptor v = spin_update[u].first;
size_t k = out_degree_no_loops(v, g);
size_t a = spin_update[u].second;
for (size_t i = 0; i < degs.size(); ++i)
int i, NK = degs.size();
#pragma omp parallel for default(shared) private(i) schedule(dynamic)
for (i = 0; i < NK; ++i)
{
size_t nk = degs[i];
double old_E = gamma*Nnnks(nk,s[v]);
global_term[nk][old_E].erase(s[v]);
if (global_term[nk][old_E].empty())
global_term[nk].erase(old_E);
old_E = gamma*Nnnks(nk,a);
global_term[nk][old_E].erase(a);
if (global_term[nk][old_E].empty())
global_term[nk].erase(old_E);
double new_E = gamma*Nnnks(nk,a);
map<double,unordered_set<size_t> >& global_term_k = global_term[nk];
unordered_set<size_t>& global_term_k_old_E = global_term_k[old_E];
unordered_set<size_t>& global_term_k_new_E = global_term_k[new_E];
global_term_k_old_E.erase(s[v]);
if (global_term_k_old_E.empty())
global_term_k.erase(old_E);
global_term_k_new_E.erase(a);
if (global_term_k_new_E.empty())
global_term_k.erase(new_E);
}
Nnnks.Update(k,s[v],a);
Ns[s[v]]--;
if (Ns[s[v]] == 0)
Ns.erase(s[v]);
Ns[a]++;
for (size_t i = 0; i < degs.size(); ++i)
#pragma omp parallel for default(shared) private(i) schedule(dynamic)
for (i = 0; i < NK; ++i)
{
size_t nk = degs[i];
global_term[nk][gamma*Nnnks(nk,s[v])].insert(s[v]);
global_term[nk][gamma*Nnnks(nk,a)].insert(a);
map<double,unordered_set<size_t> >& global_term_k = global_term[nk];
double old_E = gamma*Nnnks(nk,s[v]);
double new_E = gamma*Nnnks(nk,a);
global_term_k[old_E].insert(s[v]);
global_term_k[new_E].insert(a);
}
// update spin
......@@ -346,7 +358,7 @@ struct get_communities
size_t n_energy = 0;
for (typeof(global_term.begin()) iter = global_term.begin(); iter != global_term.end(); ++iter)
n_energy += iter->second.size();
out_str << setw(lexical_cast<string>(Nspins.first*degs.size()).size()) << n_energy;
out_str << setw(lexical_cast<string>(Nspins.first*degs.size()).size()) << n_energy << " ";
if (steepest_descent)
out_str << " (steepest descent)";
cout << out_str.str() << flush;
......@@ -367,8 +379,24 @@ struct get_communities
}
}
// get final energy
double E = 0.0;
unordered_map<size_t,vector<typename graph_traits<Graph>::vertex_descriptor> > spin_groups;
for (tie(v,v_end) = vertices(g); v != v_end; ++v)
{
typename graph_traits<Graph>::out_edge_iterator e, e_end;
for (tie(e,e_end) = out_edges(*v,g); e != e_end; ++e)
{
typename graph_traits<Graph>::vertex_descriptor t = target(*e,g);
if (s[t] == s[*v])
E -= get(weights, *e);
}
E += gamma*Nnnks(out_degree_no_loops(*v,g), s[*v]);
}
if (verbose.first)
cout << endl;
cout << " total energy: " << scientific << setprecision(20) << E << endl;
// rename spins, starting from zero
unordered_map<size_t,size_t> spins;
......@@ -377,7 +405,7 @@ struct get_communities
if (spins.find(s[*v]) == spins.end())
spins[s[*v]] = spins.size() - 1;
s[*v] = spins[s[*v]];
}
}
}
};
......@@ -420,7 +448,7 @@ public:
private:
double _p;
ManagedUnorderedMap<size_t,size_t> _Ns;
unordered_map<size_t,size_t> _Ns;
};
template <class Graph, class CommunityMap>
......@@ -458,7 +486,7 @@ public:
private:
Graph& _g;
size_t _K;
ManagedUnorderedMap<size_t,size_t> _Ks;
unordered_map<size_t,size_t> _Ks;
};
template <class Graph, class CommunityMap>
......@@ -520,20 +548,26 @@ public:
void Update(size_t k, size_t old_s, size_t s)
{
for (size_t i = 0; i < _degs.size(); ++i)
int i, NK = _degs.size();
#pragma omp parallel for default(shared) private(i) schedule(dynamic)
for (i = 0; i < NK; ++i)
{
size_t k1 = _degs[i], k2 = k;
if (_Pkk.find(k1) == _Pkk.end())
continue;
if (_Pkk.find(k1)->second.find(k2) == _Pkk.find(k1)->second.end())
continue;
_NNks[k1][old_s] -= k1*_Pkk[k1][k2] * _Nks[k2][old_s]/double(_Nk[k2]);
if (_NNks[k1][old_s] == 0.0)
_NNks[k1].erase(old_s);
if (_Nks[k2].find(s) != _Nks[k2].end())
_NNks[k1][s] -= k1*_Pkk[k1][k2] * _Nks[k2][s]/double(_Nk[k2]);
if (_NNks[k1][s] == 0.0)
_NNks[k1].erase(s);
unordered_map<size_t,double>& NNks_k1 = _NNks[k1];
double Pk1k2 = _Pkk[k1][k2];
unordered_map<size_t,size_t>& Nksk2 = _Nks[k2];
double Nk2 = _Nk[k2];
NNks_k1[old_s] -= k1*Pk1k2 * Nksk2[old_s]/Nk2;
if (NNks_k1[old_s] == 0.0)
NNks_k1.erase(old_s);
if (Nksk2.find(s) != Nksk2.end())
NNks_k1[s] -= k1*Pk1k2 * Nksk2[s]/Nk2;
if (NNks_k1[s] == 0.0)
NNks_k1.erase(s);
}
_Nks[k][old_s]--;
......@@ -541,17 +575,22 @@ public:
_Nks[k].erase(old_s);
_Nks[k][s]++;
for (size_t i = 0; i < _degs.size(); ++i)
#pragma omp parallel for default(shared) private(i) schedule(dynamic)
for (i = 0; i < NK; ++i)
{
size_t k1 = _degs[i], k2 = k;
if (_Pkk.find(k1) == _Pkk.end())
continue;
if (_Pkk.find(k1)->second.find(k2) == _Pkk.find(k1)->second.end())
continue;
_NNks[k1][old_s] += k1*_Pkk[k1][k2] * _Nks[k2][old_s]/double(_Nk[k2]);
if (_NNks[k1][old_s] == 0.0)
_NNks[k1].erase(old_s);
_NNks[k1][s] += k1*_Pkk[k1][k2] * _Nks[k2][s]/double(_Nk[k2]);
unordered_map<size_t,double>& NNks_k1 = _NNks[k1];
double Pk1k2 = _Pkk[k1][k2];
unordered_map<size_t,size_t>& Nksk2 = _Nks[k2];
double Nk2 = _Nk[k2];
NNks_k1[old_s] += k1*Pk1k2 * Nksk2[old_s]/Nk2;
if (NNks_k1[old_s] == 0.0)
NNks_k1.erase(old_s);
NNks_k1[s] += k1*Pk1k2 * Nksk2[s]/Nk2;
}
}
......@@ -570,8 +609,8 @@ private:
vector<size_t> _degs;
unordered_map<size_t,size_t> _Nk;
unordered_map<size_t,unordered_map<size_t,double> > _Pkk;
unordered_map<size_t,ManagedUnorderedMap<size_t,size_t> > _Nks;
unordered_map<size_t,ManagedUnorderedMap<size_t,double> > _NNks;
unordered_map<size_t,unordered_map<size_t,size_t> > _Nks;
unordered_map<size_t,unordered_map<size_t,double> > _NNks;
};
......
src/graph/graph_distance.cc
View file @ 0d0f4966
......@@ -70,8 +70,11 @@ struct get_distance_histogram
typename graph_traits<Graph>::vertex_iterator v2, v_end;
for (tie(v2, v_end) = vertices(g); v2 != v_end; ++v2)
if (*v2 != v && dist_map[*v2] != numeric_limits<double>::max())
{
double dist = dist_map[*v2];
#pragma omp atomic
hist[dist_map[*v2]]++;
hist[dist]++;
}
}
}
......
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment