mirrors/z3
3
0
Fork 0
mirror of https://github.com/Z3Prover/z3 synced 2025-04-15 13:28:47 +00:00
Code Activity

refactor weighted-maxsat into separate files

Browse source 
Signed-off-by: Nikolaj Bjorner <nbjorner@microsoft.com>
This commit is contained in:
Nikolaj Bjorner 2014-07-28 08:31:57 -07:00
parent 9f1b2ccfc4
commit 4ab27eff78
20 changed files with 2179 additions and 1446 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

408
src/opt/bcd2.cpp Normal file
View file

@ -0,0 +1,408 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
bcd2.cpp
Abstract:
bcd2 based MaxSAT.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#include "bcd2.h"
#include "pb_decl_plugin.h"
#include "uint_set.h"
#include "ast_pp.h"
namespace opt {
// ------------------------------------------------------
// Morgado, Heras, Marques-Silva 2013
// (initial version without model-based optimizations)
//
class bcd2 : public maxsmt_solver_base {
struct wcore {
expr* m_r;
unsigned_vector m_R;
rational m_lower;
rational m_mid;
rational m_upper;
};
typedef obj_hashtable<expr> expr_set;
pb_util pb;
expr_ref_vector m_soft_aux;
obj_map<expr, unsigned> m_relax2index; // expr |-> index
obj_map<expr, unsigned> m_soft2index; // expr |-> index
expr_ref_vector m_trail;
expr_ref_vector m_soft_constraints;
expr_set m_asm_set;
vector<wcore> m_cores;
vector<rational> m_sigmas;
rational m_den; // least common multiplier of original denominators
bool m_enable_lazy; // enable adding soft constraints lazily (called 'mgbcd2')
unsigned_vector m_lazy_soft; // soft constraints to add lazily.
void set2asms(expr_set const& set, expr_ref_vector & es) const {
es.reset();
expr_set::iterator it = set.begin(), end = set.end();
for (; it != end; ++it) {
es.push_back(m.mk_not(*it));
}
}
void bcd2_init_soft(vector<rational> const& weights, expr_ref_vector const& soft) {
// normalize weights to be integral:
m_den = rational::one();
for (unsigned i = 0; i < m_weights.size(); ++i) {
m_den = lcm(m_den, denominator(m_weights[i]));
}
if (!m_den.is_one()) {
for (unsigned i = 0; i < m_weights.size(); ++i) {
m_weights[i] = m_den*m_weights[i];
SASSERT(m_weights[i].is_int());
}
}
}
void init_bcd() {
m_trail.reset();
m_asm_set.reset();
m_cores.reset();
m_sigmas.reset();
m_lazy_soft.reset();
for (unsigned i = 0; i < m_soft.size(); ++i) {
m_sigmas.push_back(m_weights[i]);
m_soft_aux.push_back(mk_fresh());
if (m_enable_lazy) {
m_lazy_soft.push_back(i);
}
else {
enable_soft_constraint(i);
}
}
m_upper += rational(1);
}
void process_sat() {
svector<bool> assignment;
update_assignment(assignment);
if (check_lazy_soft(assignment)) {
update_sigmas();
}
}
public:
bcd2(opt_solver* s, ast_manager& m, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft):
maxsmt_solver_base(s, m, p, ws, soft),
pb(m),
m_soft_aux(m),
m_trail(m),
m_soft_constraints(m),
m_enable_lazy(true) {
bcd2_init_soft(ws, soft);
}
virtual ~bcd2() {}
virtual lbool operator()() {
expr_ref fml(m), r(m);
lbool is_sat = l_undef;
expr_ref_vector asms(m);
enable_sls();
solver::scoped_push _scope1(s());
init();
init_bcd();
if (m_cancel) {
normalize_bounds();
return l_undef;
}
process_sat();
while (m_lower < m_upper) {
IF_VERBOSE(1, verbose_stream() << "(opt.bcd2 [" << m_lower << ":" << m_upper << "])\n";);
assert_soft();
solver::scoped_push _scope2(s());
TRACE("opt", display(tout););
assert_cores();
set2asms(m_asm_set, asms);
if (m_cancel) {
normalize_bounds();
return l_undef;
}
is_sat = s().check_sat(asms.size(), asms.c_ptr());
switch(is_sat) {
case l_undef:
normalize_bounds();
return l_undef;
case l_true:
process_sat();
break;
case l_false: {
ptr_vector<expr> unsat_core;
uint_set subC, soft;
s().get_unsat_core(unsat_core);
core2indices(unsat_core, subC, soft);
SASSERT(unsat_core.size() == subC.num_elems() + soft.num_elems());
if (soft.num_elems() == 0 && subC.num_elems() == 1) {
unsigned s = *subC.begin();
wcore& c_s = m_cores[s];
c_s.m_lower = refine(c_s.m_R, c_s.m_mid);
c_s.m_mid = div(c_s.m_lower + c_s.m_upper, rational(2));
}
else {
wcore c_s;
rational delta = min_of_delta(subC);
rational lower = sum_of_lower(subC);
union_Rs(subC, c_s.m_R);
r = mk_fresh();
relax(subC, soft, c_s.m_R, delta);
c_s.m_lower = refine(c_s.m_R, lower + delta - rational(1));
c_s.m_upper = rational::one();
c_s.m_upper += sum_of_sigmas(c_s.m_R);
c_s.m_mid = div(c_s.m_lower + c_s.m_upper, rational(2));
c_s.m_r = r;
m_asm_set.insert(r);
subtract(m_cores, subC);
m_relax2index.insert(r, m_cores.size());
m_cores.push_back(c_s);
}
break;
}
}
m_lower = compute_lower();
}
normalize_bounds();
return l_true;
}
private:
void enable_soft_constraint(unsigned i) {
expr_ref fml(m);
expr* r = m_soft_aux[i].get();
m_soft2index.insert(r, i);
fml = m.mk_or(r, m_soft[i].get());
m_soft_constraints.push_back(fml);
m_asm_set.insert(r);
SASSERT(m_weights[i].is_int());
}
void assert_soft() {
for (unsigned i = 0; i < m_soft_constraints.size(); ++i) {
s().assert_expr(m_soft_constraints[i].get());
}
m_soft_constraints.reset();
}
bool check_lazy_soft(svector<bool> const& assignment) {
bool all_satisfied = true;
for (unsigned i = 0; i < m_lazy_soft.size(); ++i) {
unsigned j = m_lazy_soft[i];
if (!assignment[j]) {
enable_soft_constraint(j);
m_lazy_soft[i] = m_lazy_soft.back();
m_lazy_soft.pop_back();
--i;
all_satisfied = false;
}
}
return all_satisfied;
}
void normalize_bounds() {
m_lower /= m_den;
m_upper /= m_den;
}
expr* mk_fresh() {
expr* r = mk_fresh_bool("r");
m_trail.push_back(r);
return r;
}
void update_assignment(svector<bool>& new_assignment) {
expr_ref val(m);
rational new_upper(0);
model_ref model;
new_assignment.reset();
s().get_model(model);
for (unsigned i = 0; i < m_soft.size(); ++i) {
VERIFY(model->eval(m_soft[i].get(), val));
new_assignment.push_back(m.is_true(val));
if (!new_assignment[i]) {
new_upper += m_weights[i];
}
}
if (new_upper < m_upper) {
m_upper = new_upper;
m_model = model;
m_assignment.reset();
m_assignment.append(new_assignment);
}
}
void update_sigmas() {
for (unsigned i = 0; i < m_cores.size(); ++i) {
wcore& c_i = m_cores[i];
unsigned_vector const& R = c_i.m_R;
c_i.m_upper.reset();
for (unsigned j = 0; j < R.size(); ++j) {
unsigned r_j = R[j];
if (!m_assignment[r_j]) {
c_i.m_upper += m_weights[r_j];
m_sigmas[r_j] = m_weights[r_j];
}
else {
m_sigmas[r_j].reset();
}
}
c_i.m_mid = div(c_i.m_lower + c_i.m_upper, rational(2));
}
}
/**
* Minimum of two (positive) numbers. Zero is treated as +infinity.
*/
rational min_z(rational const& a, rational const& b) {
if (a.is_zero()) return b;
if (b.is_zero()) return a;
if (a < b) return a;
return b;
}
rational min_of_delta(uint_set const& subC) {
rational delta(0);
for (uint_set::iterator it = subC.begin(); it != subC.end(); ++it) {
unsigned j = *it;
wcore const& core = m_cores[j];
rational new_delta = rational(1) + core.m_upper - core.m_mid;
SASSERT(new_delta.is_pos());
delta = min_z(delta, new_delta);
}
return delta;
}
rational sum_of_lower(uint_set const& subC) {
rational lower(0);
for (uint_set::iterator it = subC.begin(); it != subC.end(); ++it) {
lower += m_cores[*it].m_lower;
}
return lower;
}
rational sum_of_sigmas(unsigned_vector const& R) {
rational sum(0);
for (unsigned i = 0; i < R.size(); ++i) {
sum += m_sigmas[R[i]];
}
return sum;
}
void union_Rs(uint_set const& subC, unsigned_vector& R) {
for (uint_set::iterator it = subC.begin(); it != subC.end(); ++it) {
R.append(m_cores[*it].m_R);
}
}
rational compute_lower() {
rational result(0);
for (unsigned i = 0; i < m_cores.size(); ++i) {
result += m_cores[i].m_lower;
}
return result;
}
void subtract(vector<wcore>& cores, uint_set const& subC) {
unsigned j = 0;
for (unsigned i = 0; i < cores.size(); ++i) {
if (subC.contains(i)) {
m_asm_set.remove(cores[i].m_r);
}
else {
if (j != i) {
cores[j] = cores[i];
}
++j;
}
}
cores.resize(j);
for (unsigned i = 0; i < cores.size(); ++i) {
m_relax2index.insert(cores[i].m_r, i);
}
}
void core2indices(ptr_vector<expr> const& core, uint_set& subC, uint_set& soft) {
for (unsigned i = 0; i < core.size(); ++i) {
unsigned j;
expr* a;
VERIFY(m.is_not(core[i], a));
if (m_relax2index.find(a, j)) {
subC.insert(j);
}
else {
VERIFY(m_soft2index.find(a, j));
soft.insert(j);
}
}
}
rational refine(unsigned_vector const& idx, rational v) {
return v + rational(1);
}
void relax(uint_set& subC, uint_set& soft, unsigned_vector& R, rational& delta) {
for (uint_set::iterator it = soft.begin(); it != soft.end(); ++it) {
R.push_back(*it);
delta = min_z(delta, m_weights[*it]);
m_asm_set.remove(m_soft_aux[*it].get());
}
}
void assert_cores() {
for (unsigned i = 0; i < m_cores.size(); ++i) {
assert_core(m_cores[i]);
}
}
void assert_core(wcore const& core) {
expr_ref fml(m);
vector<rational> ws;
ptr_vector<expr> rs;
rational w(0);
for (unsigned j = 0; j < core.m_R.size(); ++j) {
unsigned idx = core.m_R[j];
ws.push_back(m_weights[idx]);
w += ws.back();
rs.push_back(m_soft_aux[idx].get());
}
w.neg();
w += core.m_mid;
ws.push_back(w);
rs.push_back(core.m_r);
fml = pb.mk_le(ws.size(), ws.c_ptr(), rs.c_ptr(), core.m_mid);
s().assert_expr(fml);
}
void display(std::ostream& out) {
out << "[" << m_lower << ":" << m_upper << "]\n";
s().display(out);
out << "\n";
for (unsigned i = 0; i < m_cores.size(); ++i) {
wcore const& c = m_cores[i];
out << mk_pp(c.m_r, m) << ": ";
for (unsigned j = 0; j < c.m_R.size(); ++j) {
out << c.m_R[j] << " (" << m_sigmas[c.m_R[j]] << ") ";
}
out << "[" << c.m_lower << ":" << c.m_mid << ":" << c.m_upper << "]\n";
}
for (unsigned i = 0; i < m_soft.size(); ++i) {
out << mk_pp(m_soft[i].get(), m) << " " << m_weights[i] << "\n";
}
}
};
maxsmt_solver_base* opt::mk_bcd2(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft) {
return alloc(bcd2, s, m, p, ws, soft);
}
}

29
src/opt/bcd2.h Normal file
View file

@ -0,0 +1,29 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
bcd2.h
Abstract:
Bcd2 based MaxSAT.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#ifndef _BCD2_H_
#define _BCD2_H_
#include "maxsmt.h"
namespace opt {
maxsmt_solver_base* mk_bcd2(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
}
#endif

429
src/opt/dual_maxres.cpp Normal file
View file

@ -0,0 +1,429 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
dual_maxsres.cpp
Abstract:
MaxRes (weighted) max-sat algorithm
based on dual refinement of bounds.
MaxRes is a core-guided approach to maxsat.
DualMaxRes extends the core-guided approach by
leveraging both cores and satisfying assignments
to make progress towards a maximal satisfying assignment.
Given a (minimal) unsatisfiable core for the soft
constraints the approach works like max-res.
Given a (maximal) satisfying subset of the soft constraints
the approach updates the upper bound if the current assignment
improves the current best assignmet.
Furthermore, take the soft constraints that are complements
to the current satisfying subset.
E.g, if F are the hard constraints and
s1,...,sn, t1,..., tm are the soft clauses and
F & s1 & ... & sn is satisfiable, then the complement
of of the current satisfying subset is t1, .., tm.
Update the hard constraint:
F := F & (t1 or ... or tm)
Replace t1, .., tm by m-1 new soft clauses:
t1 & t2, t3 & (t1 or t2), t4 & (t1 or t2 or t3), ..., tn & (t1 or ... t_{n-1})
Claim:
If k of these soft clauses are satisfied, then k+1 of
the previous soft clauses are satisfied.
If k of these soft clauses are false in the satisfying assignment
for the updated F, then k of the original soft clauses are also false
under the assignment.
In summary: any assignment to the new clauses that satsfies F has the
same cost.
Claim:
If there are no satisfying assignments to F, then the current best assignment
is the optimum.
Author:
Nikolaj Bjorner (nbjorner) 2014-27-7
Notes:
--*/
#include "solver.h"
#include "maxsmt.h"
#include "dual_maxres.h"
#include "ast_pp.h"
#include "mus.h"
using namespace opt;
class dual_maxres : public maxsmt_solver_base {
expr_ref_vector m_B;
expr_ref_vector m_asms;
obj_map<expr, rational> m_asm2weight;
ptr_vector<expr> m_new_core;
mus m_mus;
expr_ref_vector m_trail;
public:
dual_maxres(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft):
maxsmt_solver_base(s, m, p, ws, soft),
m_B(m), m_asms(m),
m_mus(m_s, m),
m_trail(m)
{
}
virtual ~dual_maxres() {}
bool is_literal(expr* l) {
return
is_uninterp_const(l) ||
(m.is_not(l, l) && is_uninterp_const(l));
}
void add_soft(expr* e, rational const& w) {
TRACE("opt", tout << mk_pp(e, m) << "\n";);
expr_ref asum(m), fml(m);
if (is_literal(e)) {
asum = e;
}
else {
asum = mk_fresh_bool("soft");
fml = m.mk_iff(asum, e);
m_s->assert_expr(fml);
}
new_assumption(asum, w);
m_upper += w;
}
void new_assumption(expr* e, rational const& w) {
m_asm2weight.insert(e, w);
m_asms.push_back(e);
m_trail.push_back(e);
}
lbool operator()() {
solver::scoped_push _sc(*m_s.get());
init();
init_local();
lbool was_sat = l_false;
ptr_vector<expr> soft_compl;
while (m_lower < m_upper) {
TRACE("opt",
display_vec(tout, m_asms.size(), m_asms.c_ptr());
m_s->display(tout);
tout << "\n";
display(tout);
);
lbool is_sat = m_s->check_sat(0, 0);
if (m_cancel) {
return l_undef;
}
if (is_sat == l_true) {
was_sat = l_true;
is_sat = extend_model(soft_compl);
switch (is_sat) {
case l_undef:
break;
case l_false:
is_sat = process_unsat(soft_compl);
break;
case l_true:
is_sat = process_sat(soft_compl);
break;
}
}
switch (is_sat) {
case l_undef:
return l_undef;
case l_false:
m_lower = m_upper;
return was_sat;
case l_true:
break;
}
}
return was_sat;
}
lbool process_sat(ptr_vector<expr>& softc) {
expr_ref fml(m), tmp(m);
TRACE("opt", display_vec(tout << "softc: ", softc.size(), softc.c_ptr()););
SASSERT(!softc.empty()); // we should somehow stop if all soft are satisfied.
if (softc.empty()) {
return l_false;
}
remove_soft(softc);
rational w = split_soft(softc);
TRACE("opt", display_vec(tout << " softc: ", softc.size(), softc.c_ptr()););
dual_max_resolve(softc, w);
return l_true;
}
lbool process_unsat(ptr_vector<expr>& core) {
expr_ref fml(m);
TRACE("opt", display_vec(tout << "core: ", core.size(), core.c_ptr()););
SASSERT(!core.empty());
lbool is_sat = minimize_core(core);
SASSERT(!core.empty());
if (core.empty()) {
return l_false;
}
if (is_sat != l_true) {
return is_sat;
}
remove_soft(core);
rational w = split_soft(core);
TRACE("opt", display_vec(tout << "minimized core: ", core.size(), core.c_ptr()););
max_resolve(core, w);
m_lower += w;
IF_VERBOSE(1, verbose_stream() <<
"(opt.dual_max_res [" << m_lower << ":" << m_upper << "])\n";);
return is_sat;
}
//
// The hard constraints are satisfiable.
// Extend the current model to satisfy as many
// soft constraints as possible until either
// hitting an unsatisfiable subset of size < 1/2*#assumptions,
// or producing a maximal satisfying assignment exceeding
// number of soft constraints >= 1/2*#assumptions.
// In both cases, soft constraints that are not satisfied
// is <= 1/2*#assumptions. In this way, the new modified assumptions
// account for at most 1/2 of the current assumptions.
// The core reduction algorithms also need to take into account
// at most 1/2 of the assumptions for minimization.
//
lbool extend_model(ptr_vector<expr>& soft_compl) {
ptr_vector<expr> asms;
model_ref mdl;
expr_ref tmp(m);
m_s->get_model(mdl);
unsigned num_true = update_model(mdl, asms, soft_compl);
for (unsigned j = 0; j < m_asms.size(); ++j) {
expr* fml = m_asms[j].get();
VERIFY(mdl->eval(fml, tmp));
if (m.is_false(tmp)) {
asms.push_back(fml);
lbool is_sat = m_s->check_sat(asms.size(), asms.c_ptr());
asms.pop_back();
switch (is_sat) {
case l_false:
if (num_true*2 < m_asms.size()) {
soft_compl.reset();
m_s->get_unsat_core(soft_compl);
return l_false;
}
break;
case l_true:
m_s->get_model(mdl);
num_true = update_model(mdl, asms, soft_compl);
break;
case l_undef:
return l_undef;
}
}
}
return l_true;
}
unsigned update_model(model_ref& mdl, ptr_vector<expr>& asms, ptr_vector<expr>& soft_compl) {
expr_ref tmp(m);
asms.reset();
soft_compl.reset();
rational weight = m_lower;
unsigned num_true = 0;
for (unsigned i = 0; i < m_asms.size(); ++i) {
expr* fml = m_asms[i].get();
VERIFY(mdl->eval(fml, tmp));
SASSERT(m.is_false(tmp) || m.is_true(tmp));
if (m.is_false(tmp)) {
weight += get_weight(fml);
soft_compl.push_back(fml);
}
else {
++num_true;
asms.push_back(fml);
}
}
if (weight < m_upper) {
m_upper = weight;
m_model = mdl;
for (unsigned i = 0; i < m_soft.size(); ++i) {
expr_ref tmp(m);
VERIFY(m_model->eval(m_soft[i].get(), tmp));
m_assignment[i] = m.is_true(tmp);
}
IF_VERBOSE(1, verbose_stream() <<
"(opt.dual_max_res [" << m_lower << ":" << m_upper << "])\n";);
}
return num_true;
}
lbool minimize_core(ptr_vector<expr>& core) {
m_mus.reset();
for (unsigned i = 0; i < core.size(); ++i) {
m_mus.add_soft(core[i], 1, core.c_ptr() + i);
}
unsigned_vector mus_idx;
lbool is_sat = m_mus.get_mus(mus_idx);
if (is_sat != l_true) {
return is_sat;
}
m_new_core.reset();
for (unsigned i = 0; i < mus_idx.size(); ++i) {
m_new_core.push_back(core[mus_idx[i]]);
}
core.reset();
core.append(m_new_core);
return l_true;
}
rational get_weight(expr* e) {
return m_asm2weight.find(e);
}
//
// find the minimal weight.
// soft clauses with weight larger than the minimal weight
// are (re)added as soft clauses where the weight is updated
// to subtract the minimal weight.
//
rational split_soft(ptr_vector<expr> const& soft) {
SASSERT(!soft.empty());
rational w = get_weight(soft[0]);
for (unsigned i = 1; i < soft.size(); ++i) {
rational w2 = get_weight(soft[i]);
if (w2 < w) {
w = w2;
}
}
// add fresh soft clauses for weights that are above w.
for (unsigned i = 0; i < soft.size(); ++i) {
rational w2 = get_weight(soft[i]);
if (w2 > w) {
new_assumption(soft[i], w2 - w);
}
}
return w;
}
void display_vec(std::ostream& out, unsigned sz, expr* const* args) {
for (unsigned i = 0; i < sz; ++i) {
out << mk_pp(args[i], m) << " ";
}
out << "\n";
}
void display(std::ostream& out) {
for (unsigned i = 0; i < m_asms.size(); ++i) {
expr* a = m_asms[i].get();
out << mk_pp(a, m) << " : " << get_weight(a) << "\n";
}
}
void max_resolve(ptr_vector<expr>& core, rational const& w) {
SASSERT(!core.empty());
expr_ref fml(m), asum(m);
app_ref cls(m), d(m);
m_B.reset();
m_B.append(core.size(), core.c_ptr());
d = m.mk_true();
//
// d_0 := true
// d_i := b_{i-1} and d_{i-1} for i = 1...sz-1
// soft (b_i or !d_i)
// == (b_i or !(!b_{i-1} or d_{i-1}))
// == (b_i or b_0 & b_1 & ... & b_{i-1})
//
// Soft constraint is satisfied if previous soft constraint
// holds or if it is the first soft constraint to fail.
//
// Soundness of this rule can be established using MaxRes
//
for (unsigned i = 1; i < core.size(); ++i) {
expr* b_i = m_B[i-1].get();
expr* b_i1 = m_B[i].get();
d = m.mk_and(b_i, d);
asum = mk_fresh_bool("a");
cls = m.mk_or(b_i1, d);
fml = m.mk_iff(asum, cls);
new_assumption(asum, w);
m_s->assert_expr(fml);
}
fml = m.mk_not(m.mk_and(m_B.size(), m_B.c_ptr()));
m_s->assert_expr(fml);
}
// satc are the complements of a (maximal) satisfying assignment.
void dual_max_resolve(ptr_vector<expr>& satc, rational const& w) {
SASSERT(!satc.empty());
expr_ref fml(m), asum(m);
app_ref cls(m), d(m);
m_B.reset();
m_B.append(satc.size(), satc.c_ptr());
d = m.mk_false();
//
// d_0 := false
// d_i := b_{i-1} or d_{i-1} for i = 1...sz-1
// soft (b_i and d_i)
// == (b_i and (b_0 or b_1 or ... or b_{i-1}))
//
for (unsigned i = 1; i < satc.size(); ++i) {
expr* b_i = m_B[i-1].get();
expr* b_i1 = m_B[i].get();
d = m.mk_or(b_i, d);
asum = mk_fresh_bool("a");
cls = m.mk_and(b_i1, d);
fml = m.mk_iff(asum, cls);
new_assumption(asum, w);
m_s->assert_expr(fml);
}
fml = m.mk_or(m_B.size(), m_B.c_ptr());
m_s->assert_expr(fml);
}
void remove_soft(ptr_vector<expr> const& soft) {
for (unsigned i = 0; i < m_asms.size(); ++i) {
if (soft.contains(m_asms[i].get())) {
m_asms[i] = m_asms.back();
m_asms.pop_back();
--i;
}
}
}
virtual void set_cancel(bool f) {
maxsmt_solver_base::set_cancel(f);
m_mus.set_cancel(f);
}
void init_local() {
m_upper.reset();
m_lower.reset();
m_asm2weight.reset();
m_trail.reset();
for (unsigned i = 0; i < m_soft.size(); ++i) {
add_soft(m_soft[i].get(), m_weights[i]);
}
}
};
opt::maxsmt_solver_base* opt::mk_dual_maxres(
ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft) {
return alloc(dual_maxres, m, s, p, ws, soft);
}

32
src/opt/dual_maxres.h Normal file
View file

@ -0,0 +1,32 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
dual_maxsres.h
Abstract:
MaxRes (weighted) max-sat algorithm
based on dual refinement of bounds.
Author:
Nikolaj Bjorner (nbjorner) 2014-27-7
Notes:
--*/
#ifndef _DUAL_MAXRES_H_
#define _DUAL_MAXRES_H_
namespace opt {
maxsmt_solver_base* mk_dual_maxres(
ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
};
#endif

561
src/opt/maxhs.cpp Normal file
View file

@ -0,0 +1,561 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
maxhs.cpp
Abstract:
maxhs based MaxSAT.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#include "optsmt.h"
#include "hitting_sets.h"
#include "stopwatch.h"
#include "ast_pp.h"
#include "model_smt2_pp.h"
#include "uint_set.h"
#include "maxhs.h"
namespace opt {
class scoped_stopwatch {
double& m_time;
stopwatch m_watch;
public:
scoped_stopwatch(double& time): m_time(time) {
m_watch.start();
}
~scoped_stopwatch() {
m_watch.stop();
m_time += m_watch.get_seconds();
}
};
// ----------------------------------
// MaxSatHS+MSS
// variant of MaxSAT-HS (Algorithm 9)
// that also refines upper bound during progressive calls
// to the underlying optimization solver for the soft constraints.
//
class maxhs : public maxsmt_solver_base {
struct stats {
stats() { reset(); }
void reset() { memset(this, 0, sizeof(*this)); }
unsigned m_num_iterations;
unsigned m_num_core_reductions_success;
unsigned m_num_core_reductions_failure;
unsigned m_num_model_expansions_success;
unsigned m_num_model_expansions_failure;
double m_core_reduction_time;
double m_model_expansion_time;
double m_aux_sat_time;
double m_disjoint_cores_time;
};
hitting_sets m_hs;
expr_ref_vector m_aux; // auxiliary (indicator) variables.
obj_map<expr, unsigned> m_aux2index; // expr |-> index
unsigned_vector m_core_activity; // number of times soft constraint is used in a core.
svector<bool> m_seed; // clause selected in current model.
svector<bool> m_aux_active; // active soft clauses.
ptr_vector<expr> m_asms; // assumptions (over aux)
stats m_stats;
bool m_at_lower_bound;
public:
maxhs(opt_solver* s, ast_manager& m, params_ref& p, vector<rational> const& ws, expr_ref_vector const& soft):
maxsmt_solver_base(s, m, p, ws, soft),
m_aux(m),
m_at_lower_bound(false) {
}
virtual ~maxhs() {}
virtual void set_cancel(bool f) {
maxsmt_solver_base::set_cancel(f);
m_hs.set_cancel(f);
}
virtual void collect_statistics(statistics& st) const {
maxsmt_solver_base::collect_statistics(st);
m_hs.collect_statistics(st);
st.update("maxhs-num-iterations", m_stats.m_num_iterations);
st.update("maxhs-num-core-reductions-n", m_stats.m_num_core_reductions_failure);
st.update("maxhs-num-core-reductions-y", m_stats.m_num_core_reductions_success);
st.update("maxhs-num-model-expansions-n", m_stats.m_num_model_expansions_failure);
st.update("maxhs-num-model-expansions-y", m_stats.m_num_model_expansions_success);
st.update("maxhs-core-reduction-time", m_stats.m_core_reduction_time);
st.update("maxhs-model-expansion-time", m_stats.m_model_expansion_time);
st.update("maxhs-aux-sat-time", m_stats.m_aux_sat_time);
st.update("maxhs-disj-core-time", m_stats.m_disjoint_cores_time);
}
lbool operator()() {
ptr_vector<expr> hs;
init();
init_local();
if (!disjoint_cores(hs)) {
return l_undef;
}
seed2assumptions();
while (m_lower < m_upper) {
++m_stats.m_num_iterations;
IF_VERBOSE(1, verbose_stream() <<
"(opt.maxhs [" << m_lower << ":" << m_upper << "])\n";);
TRACE("opt", tout << "(maxhs [" << m_lower << ":" << m_upper << "])\n";);
if (m_cancel) {
return l_undef;
}
lbool core_found = generate_cores(hs);
switch(core_found) {
case l_undef:
return l_undef;
case l_true: {
lbool is_sat = next_seed();
switch(is_sat) {
case l_true:
seed2hs(false, hs);
break;
case l_false:
TRACE("opt", tout << "no more seeds\n";);
IF_VERBOSE(1, verbose_stream() << "(opt.maxhs.no-more-seeds)\n";);
m_lower = m_upper;
return l_true;
case l_undef:
return l_undef;
}
break;
}
case l_false:
IF_VERBOSE(1, verbose_stream() << "(opt.maxhs.no-more-cores)\n";);
TRACE("opt", tout << "no more cores\n";);
m_lower = m_upper;
return l_true;
}
}
return l_true;
}
private:
unsigned num_soft() const { return m_soft.size(); }
void init_local() {
unsigned sz = num_soft();
app_ref fml(m), obj(m);
expr_ref_vector sum(m);
m_asms.reset();
m_seed.reset();
m_aux.reset();
m_aux_active.reset();
m_aux2index.reset();
m_core_activity.reset();
for (unsigned i = 0; i < sz; ++i) {
bool tt = is_true(m_model, m_soft[i].get());
m_seed.push_back(tt);
m_aux. push_back(mk_fresh(m.mk_bool_sort()));
m_aux_active.push_back(false);
m_core_activity.push_back(0);
m_aux2index.insert(m_aux.back(), i);
if (tt) {
m_asms.push_back(m_aux.back());
ensure_active(i);
}
}
for (unsigned i = 0; i < m_weights.size(); ++i) {
m_hs.add_weight(m_weights[i]);
}
TRACE("opt", print_seed(tout););
}
void hs2seed(ptr_vector<expr> const& hs) {
for (unsigned i = 0; i < num_soft(); ++i) {
m_seed[i] = true;
}
for (unsigned i = 0; i < hs.size(); ++i) {
m_seed[m_aux2index.find(hs[i])] = false;
}
TRACE("opt",
print_asms(tout << "hitting set: ", hs);
print_seed(tout););
}
void seed2hs(bool pos, ptr_vector<expr>& hs) {
hs.reset();
for (unsigned i = 0; i < num_soft(); ++i) {
if (pos == m_seed[i]) {
hs.push_back(m_aux[i].get());
}
}
TRACE("opt",
print_asms(tout << "hitting set: ", hs);
print_seed(tout););
}
void seed2assumptions() {
seed2hs(true, m_asms);
}
//
// Find disjoint cores for soft constraints.
//
bool disjoint_cores(ptr_vector<expr>& hs) {
scoped_stopwatch _sw(m_stats.m_disjoint_cores_time);
m_asms.reset();
svector<bool> active(num_soft(), true);
rational lower(0);
update_assumptions(active, lower, hs);
SASSERT(lower.is_zero());
while (true) {
lbool is_sat = s().check_sat(m_asms.size(), m_asms.c_ptr());
switch (is_sat) {
case l_true:
if (lower > m_lower) {
m_lower = lower;
}
return true;
case l_false:
if (!shrink()) return false;
block_up();
update_assumptions(active, lower, hs);
break;
case l_undef:
return false;
}
}
}
void update_assumptions(svector<bool>& active, rational& lower, ptr_vector<expr>& hs) {
rational arg_min(0);
expr* e = 0;
for (unsigned i = 0; i < m_asms.size(); ++i) {
unsigned index = m_aux2index.find(m_asms[i]);
active[index] = false;
if (arg_min.is_zero() || arg_min > m_weights[index]) {
arg_min = m_weights[index];
e = m_asms[i];
}
}
if (e) {
hs.push_back(e);
lower += arg_min;
}
m_asms.reset();
for (unsigned i = 0; i < num_soft(); ++i) {
if (active[i]) {
m_asms.push_back(m_aux[i].get());
ensure_active(i);
}
}
}
//
// Auxiliary Algorithm 10 for producing cores.
//
lbool generate_cores(ptr_vector<expr>& hs) {
bool core = !m_at_lower_bound;
while (true) {
hs2seed(hs);
lbool is_sat = check_subset();
switch(is_sat) {
case l_undef:
return l_undef;
case l_true:
if (!grow()) return l_undef;
block_down();
return core?l_true:l_false;
case l_false:
core = true;
if (!shrink()) return l_undef;
block_up();
find_non_optimal_hitting_set(hs);
break;
}
}
}
struct lt_activity {
maxhs& hs;
lt_activity(maxhs& hs):hs(hs) {}
bool operator()(expr* a, expr* b) const {
unsigned w1 = hs.m_core_activity[hs.m_aux2index.find(a)];
unsigned w2 = hs.m_core_activity[hs.m_aux2index.find(b)];
return w1 < w2;
}
};
//
// produce the non-optimal hitting set by using the 10% heuristic.
// of most active cores constraints.
// m_asms contains the current core.
//
void find_non_optimal_hitting_set(ptr_vector<expr>& hs) {
std::sort(m_asms.begin(), m_asms.end(), lt_activity(*this));
for (unsigned i = m_asms.size(); i > 9*m_asms.size()/10;) {
--i;
hs.push_back(m_asms[i]);
}
}
//
// retrieve the next seed that satisfies state of hs.
// state of hs must be satisfiable before optimization is called.
//
lbool next_seed() {
scoped_stopwatch _sw(m_stats.m_aux_sat_time);
TRACE("opt", tout << "\n";);
// min c_i*(not x_i) for x_i are soft clauses.
// max c_i*x_i for x_i are soft clauses
m_at_lower_bound = false;
lbool is_sat = m_hs.compute_upper();
if (is_sat == l_true) {
is_sat = m_hs.compute_lower();
}
if (is_sat == l_true) {
m_at_lower_bound = m_hs.get_upper() == m_hs.get_lower();
if (m_hs.get_lower() > m_lower) {
m_lower = m_hs.get_lower();
}
for (unsigned i = 0; i < num_soft(); ++i) {
m_seed[i] = is_active(i) && !m_hs.get_value(i);
}
TRACE("opt", print_seed(tout););
}
return is_sat;
}
//
// check assignment returned by HS with the original
// hard constraints.
//
lbool check_subset() {
TRACE("opt", tout << "\n";);
m_asms.reset();
for (unsigned i = 0; i < num_soft(); ++i) {
if (m_seed[i]) {
m_asms.push_back(m_aux[i].get());
ensure_active(i);
}
}
return s().check_sat(m_asms.size(), m_asms.c_ptr());
}
//
// extend the current assignment to one that
// satisfies as many soft constraints as possible.
// update the upper bound based on this assignment
//
bool grow() {
scoped_stopwatch _sw(m_stats.m_model_expansion_time);
model_ref mdl;
s().get_model(mdl);
for (unsigned i = 0; i < num_soft(); ++i) {
ensure_active(i);
m_seed[i] = false;
}
for (unsigned i = 0; i < m_asms.size(); ++i) {
m_seed[m_aux2index.find(m_asms[i])] = true;
}
for (unsigned i = 0; i < num_soft(); ++i) {
if (m_seed[i]) {
// already an assumption
}
else if (is_true(mdl, m_soft[i].get())) {
m_seed[i] = true;
m_asms.push_back(m_aux[i].get());
}
else {
m_asms.push_back(m_aux[i].get());
lbool is_sat = s().check_sat(m_asms.size(), m_asms.c_ptr());
switch(is_sat) {
case l_undef:
return false;
case l_false:
++m_stats.m_num_model_expansions_failure;
m_asms.pop_back();
break;
case l_true:
++m_stats.m_num_model_expansions_success;
s().get_model(mdl);
m_seed[i] = true;
break;
}
}
}
rational upper(0);
for (unsigned i = 0; i < num_soft(); ++i) {
if (!m_seed[i]) {
upper += m_weights[i];
}
}
if (upper < m_upper) {
m_upper = upper;
m_hs.set_upper(upper);
m_model = mdl;
m_assignment.reset();
m_assignment.append(m_seed);
TRACE("opt",
tout << "new upper: " << m_upper << "\n";
model_smt2_pp(tout, m, *(mdl.get()), 0););
}
DEBUG_CODE(
for (unsigned i = 0; i < num_soft(); ++i) {
SASSERT(is_true(mdl, m_soft[i].get()) == m_seed[i]);
});
return true;
}
//
// remove soft constraints from the current core.
//
bool shrink() {
scoped_stopwatch _sw(m_stats.m_core_reduction_time);
m_asms.reset();
s().get_unsat_core(m_asms);
TRACE("opt", print_asms(tout, m_asms););
obj_map<expr, unsigned> asm2index;
for (unsigned i = 0; i < m_asms.size(); ++i) {
asm2index.insert(m_asms[i], i);
}
obj_map<expr, unsigned>::iterator it = asm2index.begin(), end = asm2index.end();
for (; it != end; ++it) {
unsigned i = it->m_value;
if (i < m_asms.size()) {
expr* tmp = m_asms[i];
expr* back = m_asms.back();
m_asms[i] = back;
m_asms.pop_back();
lbool is_sat = s().check_sat(m_asms.size(), m_asms.c_ptr());
TRACE("opt", tout << "checking: " << mk_pp(tmp, m) << ": " << is_sat << "\n";);
switch(is_sat) {
case l_true:
++m_stats.m_num_core_reductions_failure;
// put back literal into core
m_asms.push_back(back);
m_asms[i] = tmp;
break;
case l_false:
// update the core
m_asms.reset();
++m_stats.m_num_core_reductions_success;
s().get_unsat_core(m_asms);
TRACE("opt", print_asms(tout, m_asms););
update_index(asm2index);
break;
case l_undef:
return false;
}
}
}
return true;
}
void print_asms(std::ostream& out, ptr_vector<expr> const& asms) {
for (unsigned j = 0; j < asms.size(); ++j) {
out << mk_pp(asms[j], m) << " ";
}
out << "\n";
}
void print_seed(std::ostream& out) {
out << "seed: ";
for (unsigned i = 0; i < num_soft(); ++i) {
out << (m_seed[i]?"1":"0");
}
out << "\n";
}
//
// must include some literal not from asms.
// (furthermore, update upper bound constraint in HS)
//
void block_down() {
uint_set indices;
unsigned_vector c_indices;
for (unsigned i = 0; i < m_asms.size(); ++i) {
indices.insert(m_aux2index.find(m_asms[i]));
}
for (unsigned i = 0; i < num_soft(); ++i) {
if (!indices.contains(i)) {
c_indices.push_back(i);
}
}
m_hs.add_exists_false(c_indices.size(), c_indices.c_ptr());
}
// should exclude some literal from core.
void block_up() {
unsigned_vector indices;
for (unsigned i = 0; i < m_asms.size(); ++i) {
unsigned index = m_aux2index.find(m_asms[i]);
m_core_activity[index]++;
indices.push_back(index);
}
m_hs.add_exists_true(indices.size(), indices.c_ptr());
}
void update_index(obj_map<expr, unsigned>& asm2index) {
obj_map<expr, unsigned>::iterator it = asm2index.begin(), end = asm2index.end();
for (; it != end; ++it) {
it->m_value = UINT_MAX;
}
for (unsigned i = 0; i < m_asms.size(); ++i) {
asm2index.find(m_asms[i]) = i;
}
}
app_ref mk_fresh(sort* s) {
app_ref r(m);
r = m.mk_fresh_const("r", s);
m_mc->insert(r->get_decl());
return r;
}
bool is_true(model_ref& mdl, expr* e) {
expr_ref val(m);
VERIFY(mdl->eval(e, val));
return m.is_true(val);
}
bool is_active(unsigned i) const {
return m_aux_active[i];
}
void ensure_active(unsigned i) {
if (!is_active(i)) {
expr_ref fml(m);
fml = m.mk_implies(m_aux[i].get(), m_soft[i].get());
s().assert_expr(fml);
m_aux_active[i] = true;
}
}
};
maxsmt_solver_base* opt::mk_maxhs(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft) {
return alloc(maxhs, s, m, p, ws, soft);
}
}

29
src/opt/maxhs.h Normal file
View file

@ -0,0 +1,29 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
maxhs.h
Abstract:
HS-max based MaxSAT.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#ifndef _HS_MAX_H_
#define _HS_MAX_H_
#include "maxsmt.h"
namespace opt {
maxsmt_solver_base* mk_maxhs(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
}
#endif

2
src/opt/maxres.cpp
View file

@ -135,7 +135,7 @@ public:
m_lower += w; m_lower += w;
break; break;
} }
IF_VERBOSE(1, verbose_stream() << "(opt.max_res lower: " << m_lower << ")\n";); IF_VERBOSE(1, verbose_stream() << "(opt.max_res [" << m_lower << ":" << m_upper << "])\n";);
} }
return l_true; return l_true;
} }

63
src/opt/maxsls.cpp Normal file
View file

@ -0,0 +1,63 @@
/*++
Copyright (c) 2013 Microsoft Corporation
Module Name:
maxsls.cpp
Abstract:
Weighted SLS MAXSAT module
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#include "maxsls.h"
#include "ast_pp.h"
#include "model_smt2_pp.h"
namespace opt {
class sls : public maxsmt_solver_base {
public:
sls(opt_solver* s, ast_manager& m, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft):
maxsmt_solver_base(s, m, p, ws, soft) {
}
virtual ~sls() {}
lbool operator()() {
IF_VERBOSE(1, verbose_stream() << "(opt.sls)\n";);
enable_bvsat();
enable_sls();
init();
lbool is_sat = s().check_sat(0, 0);
if (is_sat == l_true) {
s().get_model(m_model);
m_upper.reset();
for (unsigned i = 0; i < m_soft.size(); ++i) {
expr_ref tmp(m);
m_model->eval(m_soft[i].get(), tmp, true);
m_assignment[i] = m.is_true(tmp);
if (!m_assignment[i]) {
m_upper += m_weights[i];
}
}
}
return is_sat;
}
};
maxsmt_solver_base* opt::mk_sls(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft) {
return alloc(sls, s, m, p, ws, soft);
}
};

36
src/opt/maxsls.h Normal file
View file

@ -0,0 +1,36 @@
/*++
Copyright (c) 2013 Microsoft Corporation
Module Name:
maxsls.h
Abstract:
Weighted SLS MAXSAT module
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
Partial, one-round SLS optimizer. Finds the first
local maximum given a resource bound and returns.
--*/
#ifndef _OPT_SLS_MAX_SAT_H_
#define _OPT_SLS_MAX_SAT_H_
#include "maxsmt.h"
namespace opt {
maxsmt_solver_base* mk_sls(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
};
#endif

17
src/opt/maxsmt.cpp
View file

@ -22,7 +22,13 @@ Notes:
#include "fu_malik.h" #include "fu_malik.h"
#include "core_maxsat.h" #include "core_maxsat.h"
#include "maxres.h" #include "maxres.h"
#include "weighted_maxsat.h" #include "dual_maxres.h"
#include "maxhs.h"
#include "bcd2.h"
#include "wpm2.h"
#include "pbmax.h"
#include "wmax.h"
#include "maxsls.h"
#include "ast_pp.h" #include "ast_pp.h"
#include "opt_params.hpp" #include "opt_params.hpp"
#include "pb_decl_plugin.h" #include "pb_decl_plugin.h"
@ -173,6 +179,9 @@ namespace opt {
else if (m_maxsat_engine == symbol("maxres")) { else if (m_maxsat_engine == symbol("maxres")) {
m_msolver = mk_maxres(m, s, m_params, m_weights, m_soft_constraints); m_msolver = mk_maxres(m, s, m_params, m_weights, m_soft_constraints);
} }
else if (m_maxsat_engine == symbol("dual-maxres")) {
m_msolver = mk_dual_maxres(m, s, m_params, m_weights, m_soft_constraints);
}
else if (m_maxsat_engine == symbol("pbmax")) { else if (m_maxsat_engine == symbol("pbmax")) {
m_msolver = mk_pbmax(m, s, m_params, m_weights, m_soft_constraints); m_msolver = mk_pbmax(m, s, m_params, m_weights, m_soft_constraints);
} }
@ -182,8 +191,8 @@ namespace opt {
else if (m_maxsat_engine == symbol("bcd2")) { else if (m_maxsat_engine == symbol("bcd2")) {
m_msolver = mk_bcd2(m, s, m_params, m_weights, m_soft_constraints); m_msolver = mk_bcd2(m, s, m_params, m_weights, m_soft_constraints);
} }
else if (m_maxsat_engine == symbol("hsmax")) { else if (m_maxsat_engine == symbol("maxhs")) {
m_msolver = mk_hsmax(m, s, m_params, m_weights, m_soft_constraints); m_msolver = mk_maxhs(m, s, m_params, m_weights, m_soft_constraints);
} }
else if (m_maxsat_engine == symbol("sls")) { else if (m_maxsat_engine == symbol("sls")) {
// NB: this is experimental one-round version of SLS // NB: this is experimental one-round version of SLS
@ -196,7 +205,7 @@ namespace opt {
m_msolver = alloc(fu_malik, m, *m_s, m_soft_constraints); m_msolver = alloc(fu_malik, m, *m_s, m_soft_constraints);
} }
else { else {
if (m_maxsat_engine != symbol::null) { if (m_maxsat_engine != symbol::null && m_maxsat_engine != symbol("wmax")) {
warning_msg("solver %s is not recognized, using default 'wmax'", warning_msg("solver %s is not recognized, using default 'wmax'",
m_maxsat_engine.str().c_str()); m_maxsat_engine.str().c_str());
} }

15
src/opt/mus.cpp
View file

@ -39,8 +39,11 @@ struct mus::imp {
expr_ref_vector m_vars; expr_ref_vector m_vars;
obj_map<expr, unsigned> m_var2idx; obj_map<expr, unsigned> m_var2idx;
volatile bool m_cancel; volatile bool m_cancel;
bool m_rmr_enabled;
imp(ref<solver>& s, ast_manager& m): m_s(s), m(m), m_cls2expr(m), m_vars(m), m_cancel(false) {} imp(ref<solver>& s, ast_manager& m):
m_s(s), m(m), m_cls2expr(m), m_vars(m), m_cancel(false),
m_rmr_enabled(false) {}
void reset() { void reset() {
m_cls2expr.reset(); m_cls2expr.reset();
@ -133,10 +136,12 @@ struct mus::imp {
assumptions.push_back(cls); assumptions.push_back(cls);
mus.push_back(cls_id); mus.push_back(cls_id);
extract_model(model); extract_model(model);
sz = core.size(); if (m_rmr_enabled) {
core.append(mus); sz = core.size();
rmr(core, mus, model); core.append(mus);
core.resize(sz); rmr(core, mus, model);
core.resize(sz);
}
break; break;
default: default:
core_exprs.reset(); core_exprs.reset();

98
src/opt/pbmax.cpp Normal file
View file

@ -0,0 +1,98 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
pbmax.cpp
Abstract:
pb based MaxSAT.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#include "pbmax.h"
#include "pb_decl_plugin.h"
#include "uint_set.h"
#include "ast_pp.h"
#include "model_smt2_pp.h"
namespace opt {
// ----------------------------------
// incrementally add pseudo-boolean
// lower bounds.
class pbmax : public maxsmt_solver_base {
public:
pbmax(opt_solver* s, ast_manager& m, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft):
maxsmt_solver_base(s, m, p, ws, soft) {
}
virtual ~pbmax() {}
lbool operator()() {
enable_bvsat();
enable_sls();
TRACE("opt", s().display(tout); tout << "\n";
for (unsigned i = 0; i < m_soft.size(); ++i) {
tout << mk_pp(m_soft[i].get(), m) << " " << m_weights[i] << "\n";
}
);
pb_util u(m);
expr_ref fml(m), val(m);
app_ref b(m);
expr_ref_vector nsoft(m);
init();
for (unsigned i = 0; i < m_soft.size(); ++i) {
nsoft.push_back(mk_not(m_soft[i].get()));
}
lbool is_sat = l_true;
while (l_true == is_sat) {
TRACE("opt", s().display(tout<<"looping\n");
model_smt2_pp(tout << "\n", m, *(m_model.get()), 0););
m_upper.reset();
for (unsigned i = 0; i < m_soft.size(); ++i) {
VERIFY(m_model->eval(nsoft[i].get(), val));
m_assignment[i] = !m.is_true(val);
if (!m_assignment[i]) {
m_upper += m_weights[i];
}
}
IF_VERBOSE(1, verbose_stream() << "(opt.pb [" << m_lower << ":" << m_upper << "])\n";);
TRACE("opt", tout << "new upper: " << m_upper << "\n";);
fml = u.mk_lt(nsoft.size(), m_weights.c_ptr(), nsoft.c_ptr(), m_upper);
solver::scoped_push _scope2(s());
s().assert_expr(fml);
is_sat = s().check_sat(0,0);
if (m_cancel) {
is_sat = l_undef;
}
if (is_sat == l_true) {
s().get_model(m_model);
}
}
if (is_sat == l_false) {
is_sat = l_true;
m_lower = m_upper;
}
TRACE("opt", tout << "lower: " << m_lower << "\n";);
return is_sat;
}
};
maxsmt_solver_base* opt::mk_pbmax(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft) {
return alloc(pbmax, s, m, p, ws, soft);
}
}

30
src/opt/pbmax.h Normal file
View file

@ -0,0 +1,30 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
pbmax.h
Abstract:
MaxSAT based on pb theory.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#ifndef _PBMAX_H_
#define _PBMAX_H_
#include "maxsmt.h"
namespace opt {
maxsmt_solver_base* mk_pbmax(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
}
#endif

1384
src/opt/weighted_maxsat.cpp
View file

File diff suppressed because it is too large Load diff

51
src/opt/weighted_maxsat.h
View file

@ -1,51 +0,0 @@
/*++
Copyright (c) 2013 Microsoft Corporation
Module Name:
weighted_maxsat.h
Abstract:
Weighted MAXSAT module
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
Takes solver with hard constraints added.
Returns a maximal satisfying subset of weighted soft_constraints
that are still consistent with the solver state.
--*/
#ifndef _OPT_WEIGHTED_MAX_SAT_H_
#define _OPT_WEIGHTED_MAX_SAT_H_
#include "opt_solver.h"
#include "maxsmt.h"
namespace opt {
maxsmt_solver_base* mk_bcd2(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
maxsmt_solver_base* mk_hsmax(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
maxsmt_solver_base* mk_pbmax(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
maxsmt_solver_base* mk_wpm2(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
maxsmt_solver_base* mk_sls(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
maxsmt_solver_base* mk_wmax(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
};
#endif

130
src/opt/wmax.cpp Normal file
View file

@ -0,0 +1,130 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
wmax.cpp
Abstract:
Theory based MaxSAT.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#include "wmax.h"
#include "uint_set.h"
#include "ast_pp.h"
#include "model_smt2_pp.h"
#include "smt_theory.h"
#include "smt_context.h"
#include "theory_wmaxsat.h"
namespace opt {
class maxsmt_solver_wbase : public maxsmt_solver_base {
smt::context& ctx;
public:
maxsmt_solver_wbase(opt_solver* s, ast_manager& m, smt::context& ctx, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft):
maxsmt_solver_base(s, m, p, ws, soft), ctx(ctx) {}
~maxsmt_solver_wbase() {}
class scoped_ensure_theory {
smt::theory_wmaxsat* m_wth;
public:
scoped_ensure_theory(maxsmt_solver_wbase& s) {
m_wth = s.ensure_theory();
}
~scoped_ensure_theory() {
m_wth->reset();
}
smt::theory_wmaxsat& operator()() { return *m_wth; }
};
smt::theory_wmaxsat* ensure_theory() {
smt::theory_wmaxsat* wth = get_theory();
if (wth) {
wth->reset();
}
else {
wth = alloc(smt::theory_wmaxsat, m, m_mc);
ctx.register_plugin(wth);
}
return wth;
}
smt::theory_wmaxsat* get_theory() const {
smt::theory_id th_id = m.get_family_id("weighted_maxsat");
smt::theory* th = ctx.get_theory(th_id);
if (th) {
return dynamic_cast<smt::theory_wmaxsat*>(th);
}
else {
return 0;
}
}
};
// ----------------------------------------------------------
// weighted max-sat using a custom theory solver for max-sat.
// NB. it is quite similar to pseudo-Boolean propagation.
class wmax : public maxsmt_solver_wbase {
public:
wmax(opt_solver* s, ast_manager& m, smt::context& ctx, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft):
maxsmt_solver_wbase(s, m, ctx, p, ws, soft) {}
virtual ~wmax() {}
lbool operator()() {
TRACE("opt", tout << "weighted maxsat\n";);
scoped_ensure_theory wth(*this);
solver::scoped_push _s(s());
lbool is_sat = l_true;
bool was_sat = false;
for (unsigned i = 0; i < m_soft.size(); ++i) {
wth().assert_weighted(m_soft[i].get(), m_weights[i]);
}
solver::scoped_push __s(s());
while (l_true == is_sat) {
IF_VERBOSE(1, verbose_stream() << "(opt.wmax [" << m_lower << ":" << m_upper << "])\n";);
is_sat = s().check_sat(0,0);
if (m_cancel) {
is_sat = l_undef;
}
if (is_sat == l_true) {
if (wth().is_optimal()) {
m_upper = wth().get_min_cost();
s().get_model(m_model);
}
expr_ref fml = wth().mk_block();
s().assert_expr(fml);
was_sat = true;
}
}
if (was_sat) {
wth().get_assignment(m_assignment);
}
if (is_sat == l_false && was_sat) {
is_sat = l_true;
}
m_upper = wth().get_min_cost();
if (is_sat == l_true) {
m_lower = m_upper;
}
TRACE("opt", tout << "min cost: " << m_upper << "\n";);
return is_sat;
}
};
maxsmt_solver_base* opt::mk_wmax(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft) {
return alloc(wmax, s, m, s->get_context(), p, ws, soft);
}
}

30
src/opt/wmax.h Normal file
View file

@ -0,0 +1,30 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
wmax.h
Abstract:
Theory Solver based MaxSAT.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#ifndef _WMAX_H_
#define _WMAX_H_
#include "maxsmt.h"
namespace opt {
maxsmt_solver_base* mk_wmax(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
}
#endif

251
src/opt/wpm2.cpp Normal file
View file

@ -0,0 +1,251 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
wpm2.cpp
Abstract:
wpn2 based MaxSAT.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#include "wpm2.h"
#include "pbmax.h"
#include "pb_decl_plugin.h"
#include "uint_set.h"
#include "ast_pp.h"
#include "model_smt2_pp.h"
namespace opt {
// ------------------------------------------------------
// AAAI 2010
class wpm2 : public maxsmt_solver_base {
scoped_ptr<maxsmt_solver_base> maxs;
public:
wpm2(opt_solver* s, ast_manager& m, maxsmt_solver_base* _maxs, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft):
maxsmt_solver_base(s, m, p, ws, soft), maxs(_maxs) {
}
virtual ~wpm2() {}
lbool operator()() {
enable_sls();
IF_VERBOSE(1, verbose_stream() << "(opt.wpm2)\n";);
solver::scoped_push _s(s());
pb_util u(m);
app_ref fml(m), a(m), b(m), c(m);
expr_ref val(m);
expr_ref_vector block(m), ans(m), al(m), am(m);
obj_map<expr, unsigned> ans_index;
vector<rational> amk;
vector<uint_set> sc;
init();
for (unsigned i = 0; i < m_soft.size(); ++i) {
rational w = m_weights[i];
b = mk_fresh_bool("b");
block.push_back(b);
expr* bb = b;
a = mk_fresh_bool("a");
ans.push_back(a);
ans_index.insert(a, i);
fml = m.mk_or(m_soft[i].get(), b, m.mk_not(a));
s().assert_expr(fml);
c = mk_fresh_bool("c");
fml = m.mk_implies(c, u.mk_le(1,&w,&bb,rational(0)));
s().assert_expr(fml);
sc.push_back(uint_set());
sc.back().insert(i);
am.push_back(c);
amk.push_back(rational(0));
}
while (true) {
expr_ref_vector asms(m);
ptr_vector<expr> core;
asms.append(ans);
asms.append(am);
lbool is_sat = s().check_sat(asms.size(), asms.c_ptr());
TRACE("opt",
tout << "\nassumptions: ";
for (unsigned i = 0; i < asms.size(); ++i) {
tout << mk_pp(asms[i].get(), m) << " ";
}
tout << "\n" << is_sat << "\n";
tout << "upper: " << m_upper << "\n";
tout << "lower: " << m_lower << "\n";
if (is_sat == l_true) {
model_ref mdl;
s().get_model(mdl);
model_smt2_pp(tout, m, *(mdl.get()), 0);
});
if (m_cancel && is_sat != l_false) {
is_sat = l_undef;
}
if (is_sat == l_true) {
m_upper = m_lower;
s().get_model(m_model);
for (unsigned i = 0; i < block.size(); ++i) {
VERIFY(m_model->eval(m_soft[i].get(), val));
TRACE("opt", tout << mk_pp(block[i].get(), m) << " " << val << "\n";);
m_assignment[i] = m.is_true(val);
}
}
if (is_sat != l_false) {
return is_sat;
}
s().get_unsat_core(core);
if (core.empty()) {
return l_false;
}
TRACE("opt",
tout << "core: ";
for (unsigned i = 0; i < core.size(); ++i) {
tout << mk_pp(core[i],m) << " ";
}
tout << "\n";);
uint_set A;
for (unsigned i = 0; i < core.size(); ++i) {
unsigned j;
if (ans_index.find(core[i], j)) {
A.insert(j);
}
}
if (A.empty()) {
return l_false;
}
uint_set B;
rational k(0);
rational old_lower(m_lower);
for (unsigned i = 0; i < sc.size(); ++i) {
uint_set t(sc[i]);
t &= A;
if (!t.empty()) {
B |= sc[i];
k += amk[i];
m_lower -= amk[i];
sc[i] = sc.back();
sc.pop_back();
am[i] = am.back();
am.pop_back();
amk[i] = amk.back();
amk.pop_back();
--i;
}
}
vector<rational> ws;
expr_ref_vector bs(m);
for (unsigned i = 0; i < m_soft.size(); ++i) {
if (B.contains(i)) {
ws.push_back(m_weights[i]);
bs.push_back(block[i].get());
}
}
TRACE("opt", tout << "at most bound: " << k << "\n";);
is_sat = new_bound(al, ws, bs, k);
if (is_sat != l_true) {
return is_sat;
}
m_lower += k;
SASSERT(m_lower > old_lower);
TRACE("opt", tout << "new bound: " << m_lower << "\n";);
expr_ref B_le_k(m), B_ge_k(m);
B_le_k = u.mk_le(ws.size(), ws.c_ptr(), bs.c_ptr(), k);
B_ge_k = u.mk_ge(ws.size(), ws.c_ptr(), bs.c_ptr(), k);
s().assert_expr(B_ge_k);
al.push_back(B_ge_k);
IF_VERBOSE(1, verbose_stream() << "(opt.wpm2 [" << m_lower << ":" << m_upper << "])\n";);
IF_VERBOSE(2, verbose_stream() << "New lower bound: " << B_ge_k << "\n";);
c = mk_fresh_bool("c");
fml = m.mk_implies(c, B_le_k);
s().assert_expr(fml);
sc.push_back(B);
am.push_back(c);
amk.push_back(k);
}
}
virtual void set_cancel(bool f) {
maxsmt_solver_base::set_cancel(f);
maxs->set_cancel(f);
}
virtual void collect_statistics(statistics& st) const {
maxsmt_solver_base::collect_statistics(st);
maxs->collect_statistics(st);
}
private:
lbool new_bound(expr_ref_vector const& al,
vector<rational> const& ws,
expr_ref_vector const& bs,
rational& k) {
pb_util u(m);
expr_ref_vector al2(m);
al2.append(al);
// w_j*b_j > k
al2.push_back(m.mk_not(u.mk_le(ws.size(), ws.c_ptr(), bs.c_ptr(), k)));
return bound(al2, ws, bs, k);
}
//
// minimal k, such that al & w_j*b_j >= k is sat
// minimal k, such that al & 3*x + 4*y >= k is sat
// minimal k, such that al & (or (not x) w3) & (or (not y) w4)
//
lbool bound(expr_ref_vector const& al,
vector<rational> const& ws,
expr_ref_vector const& bs,
rational& k) {
expr_ref_vector nbs(m);
opt_solver::scoped_push _sc(maxs->s());
for (unsigned i = 0; i < al.size(); ++i) {
maxs->add_hard(al[i]);
}
for (unsigned i = 0; i < bs.size(); ++i) {
nbs.push_back(mk_not(bs[i]));
}
TRACE("opt",
maxs->s().display(tout);
tout << "\n";
for (unsigned i = 0; i < bs.size(); ++i) {
tout << mk_pp(bs[i], m) << " " << ws[i] << "\n";
});
maxs->init_soft(ws, nbs);
lbool is_sat = maxs->s().check_sat(0,0);
if (is_sat == l_true) {
maxs->set_model();
is_sat = (*maxs)();
}
SASSERT(maxs->get_lower() > k);
k = maxs->get_lower();
return is_sat;
}
};
maxsmt_solver_base* opt::mk_wpm2(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft) {
ref<opt_solver> s0 = alloc(opt_solver, m, p, symbol());
// initialize model.
s0->check_sat(0,0);
maxsmt_solver_base* s2 = mk_pbmax(m, s0.get(), p, ws, soft);
return alloc(wpm2, s, m, s2, p, ws, soft);
}
}

29
src/opt/wpm2.h Normal file
View file

@ -0,0 +1,29 @@
/*++
Copyright (c) 2014 Microsoft Corporation
Module Name:
wpm2.h
Abstract:
Wpm2 based MaxSAT.
Author:
Nikolaj Bjorner (nbjorner) 2014-4-17
Notes:
--*/
#ifndef _WPM2_H_
#define _WPM2_H_
#include "maxsmt.h"
namespace opt {
maxsmt_solver_base* mk_wpm2(ast_manager& m, opt_solver* s, params_ref& p,
vector<rational> const& ws, expr_ref_vector const& soft);
}
#endif

1
src/smt/theory_wmaxsat.cpp
View file

@ -216,7 +216,6 @@ expr_ref theory_wmaxsat::mk_block() {
scoped_mpq q(mgr); scoped_mpq q(mgr);
mgr.set(q, m_zmin_cost, m_den.to_mpq().numerator()); mgr.set(q, m_zmin_cost, m_den.to_mpq().numerator());
rational rw = rational(q); rational rw = rational(q);
IF_VERBOSE(1, verbose_stream() << "(wmaxsat with upper bound: " << rw << ")\n";);
m_zmin_cost = weight; m_zmin_cost = weight;
m_found_optimal = true; m_found_optimal = true;
m_cost_save.reset(); m_cost_save.reset();