alarsyo/spot
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity

* src/tgba/tgbaexplicit.cc, src/tgba/tgbaexplicit.hh

Browse source 
(tgba_explicit::create_transition(state*, const state*)): New function.
* src/tgbaalgos/randomgraph.cc, src/tgbaalgos/randomgraph.hh:
(random_graph): Revamp the algorithm to call rand() less often.
* src/tgbatest/randtgba.cc: Add option -0 to easy profiling.
This commit is contained in:
Alexandre Duret-Lutz 2004-12-08 15:39:15 +00:00
parent 7917841fbe
commit 688587d700
6 changed files with 129 additions and 56 deletions
Download patch file Download diff file Expand all files Collapse all files

132
src/tgbaalgos/randomgraph.cc
View file

@ -27,6 +27,8 @@
#include <sstream>
#include <list>
#include <set>
#include <iterator>
#include "misc/hash.hh"
namespace spot
{
@ -51,14 +53,29 @@ namespace spot
void
random_labels(tgba_explicit* aut,
const std::string& src, const std::string& dest,
const std::list<int>& props, float t,
tgba_explicit::state* src, const tgba_explicit::state* dest,
int* props, int props_n, float t,
const std::list<bdd>& accs, float a)
{
int val = 0;
int size = 0;
bdd p = bddtrue;
for (std::list<int>::const_iterator i = props.begin();
i != props.end(); ++i)
p &= (drand() < t ? bdd_ithvar : bdd_nithvar)(*i);
while (props_n)
{
if (size == 8 * sizeof(int))
{
p &= bdd_ibuildcube(val, size, props);
props += size;
val = 0;
size = 0;
}
val <<= 1;
val |= (drand() < t);
++size;
--props_n;
}
if (size > 0)
p &= bdd_ibuildcube(val, size, props);
bdd ac = bddfalse;
for (std::list<bdd>::const_iterator i = accs.begin();
@ -80,10 +97,15 @@ namespace spot
{
tgba_explicit* res = new tgba_explicit(dict);
std::list<int> props;
int props_n = ap->size();
int* props = new int[props_n];
int pi = 0;
for (ltl::atomic_prop_set::const_iterator i = ap->begin();
i != ap->end(); ++i)
props.push_back(dict->register_proposition(*i, res));
props[pi++] = dict->register_proposition(*i, res);
std::vector<tgba_explicit::state*> states(n);
std::list<bdd> accs;
bdd allneg = bddtrue;
@ -99,56 +121,82 @@ namespace spot
for (std::list<bdd>::iterator i = accs.begin(); i != accs.end(); ++i)
*i &= bdd_exist(allneg, *i);
typedef std::set<std::string> node_set;
// Using Sgi::hash_set instead of std::set for these sets is 3
// times slower (tested on a 50000 nodes example).
typedef std::set<tgba_explicit::state*> node_set;
node_set nodes_to_process;
node_set unreachable_nodes;
nodes_to_process.insert(st(0));
nodes_to_process.insert(states[0] = res->add_state(st(0)));
for (int i = 1; i < n; ++i)
unreachable_nodes.insert(st(i));
unreachable_nodes.insert(states[i] = res->add_state(st(i)));
// We want to connect each node to a number of successors between
// 1 and n (with probability d). This follow
barand<nrand> bin(n - 1, d);
while (!nodes_to_process.empty())
{
std::string src = *nodes_to_process.begin();
tgba_explicit::state* src = *nodes_to_process.begin();
nodes_to_process.erase(nodes_to_process.begin());
if (!unreachable_nodes.empty())
// Choose a random number of successors (at least one), using
// a binomial distribution.
int nsucc = 1 + bin.rand();
// Connect to NSUCC randomly chosen successors. We want at
// least one unreachable successors among these if there are
// some.
bool saw_unreachable = false;
int possibilities = n;
while (nsucc--)
{
// Pick a random unreachable node.
int index = mrand(unreachable_nodes.size());
node_set::const_iterator i;
for (i = unreachable_nodes.begin(); index; ++i, --index)
assert(i != unreachable_nodes.end());
// Link it from src.
random_labels(res, src, *i, props, t, accs, a);
nodes_to_process.insert(*i);
unreachable_nodes.erase(i);
}
// Randomly link node to another node (including itself).
for (int i = 0; i < n; ++i)
{
if (drand() >= d)
continue;
std::string dest = st(i);
random_labels(res, src, dest, props, t, accs, a);
node_set::iterator j = unreachable_nodes.find(dest);
if (j != unreachable_nodes.end())
if (nsucc == 0
&& !saw_unreachable
&& !unreachable_nodes.empty())
{
nodes_to_process.insert(dest);
unreachable_nodes.erase(j);
// Pick a random unreachable node.
int index = mrand(unreachable_nodes.size());
node_set::const_iterator i = unreachable_nodes.begin();
std::advance(i, index);
// Link it from src.
random_labels(res, src, *i, props, props_n, t, accs, a);
nodes_to_process.insert(*i);
unreachable_nodes.erase(i);
break;
}
else
{
// Pick a random node.
int index = mrand(possibilities--);
tgba_explicit::state* dest = states[index];
// Permute the state with states[possibilities], so we
// cannot pick it again.
states[index] = states[possibilities];
states[possibilities] = dest;
random_labels(res, src, dest, props, props_n, t, accs, a);
node_set::iterator j = unreachable_nodes.find(dest);
if (j != unreachable_nodes.end())
{
nodes_to_process.insert(dest);
unreachable_nodes.erase(j);
saw_unreachable = true;
}
}
}
// Avoid dead ends.
if (res->add_state(src)->empty())
random_labels(res, src, src, props, t, accs, a);
// The node must have at least one successor.
assert(!src->empty());
}
// All nodes must be reachable.
assert(unreachable_nodes.empty());
delete[] props;
return res;
}

11
src/tgbaalgos/randomgraph.hh
View file

@ -65,8 +65,15 @@ namespace spot
/// note = {Reprint of Master's thesis}
/// }
/// \endverbatim
/// The only difference is that labels are on transitions, and that
/// acceptance conditions are handled. Otherwise the logic is the same.
///
/// Although the intent is similar, there are some differences with
/// between the above published algorithm and this implementation .
/// First labels are on transitions, and acceptance conditions are
/// generated too. Second, the number of successors of a node is
/// chosen in \f$[1,n]\f$ following a normal distribution with mean
/// \f$1+(n-1)d\f$ and variance \f$(n-1)d(1-d)\f$. (This is less
/// accurate, but faster than considering all possible \a n
/// successors one by one.)
tgba*
random_graph(int n, float d,
const ltl::atomic_prop_set* ap, bdd_dict* dict,