capd::interval formal::evaluate_pha(int min_depth, int max_depth)
{
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Obtaining partition of domain of continuous random parameters";
// getting partition of domain of continuous random variables
std::vector<box> init_rv_partition = measure::get_rv_partition();
// getting domain of continuous random variables
box rv_domain = measure::bounds::get_rv_domain();
// here we start with entire domain instead of partition
if (!global_config.partition_prob) {
init_rv_partition.clear();
init_rv_partition.push_back(rv_domain);
}
// getting partition of domain of discrete random variables
std::vector<box> dd_partition = measure::get_dd_partition();
if (dd_partition.empty()) {
dd_partition.push_back(box());
}
capd::interval probability(0, 1);
// checking if there are any continuous random variables
CLOG_IF(global_config.verbose, INFO, "algorithm") << "P = " << probability;
// generating all paths of lengths [min_depth, max_depth]
std::vector<std::vector<pdrh::mode *>> paths = pdrh::get_all_paths(min_depth, max_depth);
//resulting probability
capd::interval res_prob(0.0);
// evaluating boxes
//#pragma omp parallel for
for (box dd : dd_partition)
{
if (pdrh::rv_map.size() > 0) probability = capd::interval(0, 2 - measure::p_measure(rv_domain, global_config.precision_prob).leftBound());
vector<box> rv_partition = init_rv_partition;
std::vector<box> rv_stack;
while (capd::intervals::width(probability) > global_config.precision_prob)
{
// sorting boxes by probability value
if (global_config.sort_rv_flag)
{
CLOG_IF(global_config.verbose, INFO, "algorithm")
<< "Sorting the partition of domain of continuous random parameters";
sort(rv_partition.begin(), rv_partition.end(), measure::compare_boxes_by_p_measure);
}
//for(box rv : rv_partition)
#pragma omp parallel for
for (size_t i = 0; i < rv_partition.size(); i++)
{
box rv = rv_partition.at(i);
// calculating probability measure of the box
// initially p_box = [1.0, 1.0]
CLOG_IF(global_config.verbose, INFO, "algorithm") << "====================";
capd::interval p_box(1);
if (!dd.empty())
{
p_box *= measure::p_dd_measure(dd);
CLOG_IF(global_config.verbose, INFO, "algorithm") << "dd_box: " << dd;
}
if (!rv.empty())
{
p_box *= measure::p_measure(rv, global_config.precision_prob);
CLOG_IF(global_config.verbose, INFO, "algorithm") << "rv_box: " << rv;
}
CLOG_IF(global_config.verbose, INFO, "algorithm") << "p_box: " << p_box;
// evaluating boxes
std::vector<box> boxes{dd, rv};
// undetermined answers counter
int undet_counter = 0;
// unsat counter
int unsat_counter = 0;
// sat flag
bool sat_flag = false;
// evaluating all paths for all dd and rv
//cout << "Before evaluate loop " << omp_get_thread_num() << endl;
for (std::vector<pdrh::mode *> path : paths) {
std::string solver_opt;
std::stringstream p_stream;
for (pdrh::mode *m : path) {
p_stream << m->id << " ";
}
// removing trailing whitespace
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Path: " << p_stream.str().substr(0,
p_stream.str().find_last_of(
" "));
std::stringstream s;
// changing solver precision
#pragma omp critical
{
solver_opt = global_config.solver_opt;
s << solver_opt << " --precision " <<
measure::volume(rv).leftBound() * global_config.solver_precision_ratio;
global_config.solver_opt = s.str();
}
int res = decision_procedure::evaluate(path, boxes, global_config.solver_bin, s.str());
// setting old precision
#pragma omp critical
{
global_config.solver_opt = solver_opt;
switch (res)
{
case decision_procedure::SAT:
if (p_box.leftBound() > 0)
{
probability = capd::interval(probability.leftBound() + p_box.leftBound(),
probability.rightBound());
}
CLOG_IF(global_config.verbose, INFO, "algorithm") << "SAT";
CLOG_IF(global_config.verbose, INFO, "algorithm") << "P = " << probability;
/*
if(capd::intervals::width(probability) <= global_config.precision_prob)
{
return probability;
}
*/
sat_flag = true;
break;
case decision_procedure::UNSAT:
CLOG_IF(global_config.verbose, INFO, "algorithm") << "UNSAT";
unsat_counter++;
break;
case decision_procedure::UNDET:
CLOG_IF(global_config.verbose, INFO, "algorithm") << "UNDET";
undet_counter++;
break;
case decision_procedure::ERROR:
CLOG(ERROR, "algorithm") << "Error occurred while calling the solver";
exit(EXIT_FAILURE);
default:
break;
}
}
// breaking out of the loop if a sat path was found
if (sat_flag) break;
}
#pragma omp critical
{
// checking if there are no sat answers
if (!sat_flag)
{
// if the box is undetermined on either path
if (undet_counter > 0)
{
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Bisect " << rv;
std::vector<box> rv_bisect = box_factory::bisect(rv);
rv_stack.insert(rv_stack.end(), rv_bisect.begin(), rv_bisect.end());
}
// if the box is unsat for all paths
else if (unsat_counter == paths.size())
{
if (p_box.rightBound() > 0)
{
probability = capd::interval(probability.leftBound(),
probability.rightBound() - p_box.leftBound());
}
CLOG_IF(global_config.verbose, INFO, "algorithm") << "P = " << probability;
}
}
};
}
// copying the bisected boxes from the stack to partition
rv_partition = rv_stack;
rv_stack.clear();
// breaking out of the loop if there are no continuous random variables
if (pdrh::rv_map.size() == 0)
{
rv_partition.push_back(box());
break;
}
}
if (!dd.empty())
{
capd::interval dd_measure = measure::p_dd_measure(dd);
res_prob += probability * dd_measure;
CLOG_IF(global_config.verbose, INFO, "algorithm") << "P(" << dd << ") = " << probability * dd_measure;
}
else
{
res_prob = probability;
}
}
return res_prob;
}
void BD_Factory::add_discretized_ctr(double eps_g) {
std::vector<double>* samples;
switch (problem) {
case LBD : samples = sip.LBD_samples; break;
case UBD : samples = sip.UBD_samples; break;
case ORA : samples = sip.ORA_samples; break;
}
// For each constraint (even non-quantified)
for (int c=0; c<sip.sys.nb_ctr; c++) {
//cout << "Constraint: " << sip.sys.ctrs[c] << endl;
// build an initial set of (all) parameters values.
// (includes those not involved in the constraint but these
// values will be ignored anyway)
Vector p_box(sip.p);
for (int j=0; j<sip.p; j++)
p_box[j]=samples[j].front();
// push this initial combination in a list
list<Vector> p_boxes;
//cout << " add p_box:" << p_box << endl;
p_boxes.push_back(p_box);
// We then create all combinations of LBD/UBD_samples for the parameters
// involved in the constraint c, via a recursive Cartesian product
// with all previously existing combinations.
for (int j=0; j<sip.p; j++) {
//cout << " param n°" << j << "=" << sip.varset.param(j);
if (sip.sys.ctrs[c].f.used(sip.varset.param(j))) {
//cout << " **used**" << endl;
// for each box already inside the list...
for (list<Vector>::iterator it2=p_boxes.begin(); it2!=p_boxes.end(); it2=p_boxes.erase(it2)) {
p_box=*it2;
// ... we instantiate the domain of the parameter to each of its current
// sampled values
for (vector<double>::iterator it=samples[j].begin(); it!=samples[j].end(); it++) {
p_box[j]=*it;
p_boxes.push_front(p_box); // pushed at the beginning -> no impact for the enclosing loop
//cout << " add p_box:" << p_box << endl;
}
}
} else {
//cout << " (unused)" << endl;
}
}
// all the constraints generated with constraint n°c
// (stored in an array for cleanup)
Array<const ExprNode> exprs_ctr(p_boxes.size());
int i=0;
// We generate one constraint for each p_box
for (list<Vector>::iterator it=p_boxes.begin(); it!=p_boxes.end(); it++) {
// transform the "flat" box into a list of domains
// (which is necessary for composition of functions expression)
load(sip.p_domain, *it);
Array<const ExprNode> new_args(sip.n_arg+sip.p_arg);
int I=0;
int J=0;
for (int K=0; K<sip.n_arg+sip.p_arg; K++) {
if (!sip.is_param[K])
// a variable x in the original system is a variable in
// the LBD/UBD problem
new_args.set_ref(K,new_vars[I++]);
else
// a parameter becomes a constant
// (note: can have a reference because the expression is copied afterward)
new_args.set_ref(K,ExprConstant::new_(sip.p_domain[J++],true));
}
const ExprNode* expr_ctr_tmp=&sip.sys.ctrs[c].f(new_args);
if (problem==UBD) {
// In the UBD problem, shift the constraint with eps_g
const ExprConstant& eps_node=ExprConstant::new_scalar(sip.sys.ctrs[c].op==LEQ ? eps_g : -eps_g);
expr_ctr_tmp = & (*expr_ctr_tmp + eps_node);
} else if (problem==ORA) {
// in the ORA problem, shift the constraint with eta
const ExprNode& eta=new_vars[sip.n_arg];
const ExprNode& eta_node=sip.sys.ctrs[c].op==LEQ ? eta : -eta;
expr_ctr_tmp = & (*expr_ctr_tmp + eta_node);
}
// cleanup the constants created which
// do not appear in the constraint expression
for (int K=0; K<sip.n_arg+sip.p_arg; K++) {
if (sip.is_param[K] && new_args[K].fathers.is_empty()) delete &new_args[K];
}
const ExprNode& expr_ctr=expr_ctr_tmp->simplify();
ExprCtr ctr(expr_ctr, sip.sys.ctrs[c].op);
//cout << " generated constraint:" << ctr << endl;
add_ctr(ctr);
exprs_ctr.set_ref(i++,expr_ctr);
}
cleanup(exprs_ctr,false);
f_ctrs_copy.clone.clean();
//cout << endl;
}
}
std::map<box, capd::interval> formal::evaluate_npha(int min_depth, int max_depth)
{
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Obtaining domain of nondeterministic parameters";
// getting parameter domain
box nd_domain = pdrh2box::get_nondet_domain();
// initially partition is the entire parameter domain
std::vector<box> nd_partition{nd_domain};
// if flag is enabled the domain is partitioned up to precision_nondet
if (global_config.partition_nondet) {
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Obtaining partition of nondeterministic domain";
nd_partition.clear();
//nd_partition = box_factory::partition(nd_domain, global_config.precision_nondet);
nd_partition = box_factory::partition(nd_domain, global_config.partition_nondet_map);
}
// getting partition of domain of continuous random variables
CLOG_IF(global_config.verbose, INFO, "algorithm")
<< "Obtaining partition of domain of continuous random parameters";
std::vector<box> rv_partition = measure::get_rv_partition();
if (rv_partition.empty()) {
rv_partition.push_back(box());
}
// getting domain of continuous random variables
box rv_domain = measure::bounds::get_rv_domain();
// here we start with entire domain instead of partition
if (!global_config.partition_prob) {
rv_partition.clear();
rv_partition.push_back(rv_domain);
}
// performing extra partition if the probability partition map is defined
else
{
//cout << "rv partition size before extra partition: " << rv_partition.size() << endl;
vector<box> tmp_vector;
for (box b : rv_partition)
{
vector<box> extra_partition = box_factory::partition(b, global_config.partition_prob_map);
tmp_vector.insert(tmp_vector.cend(), extra_partition.cbegin(), extra_partition.cend());
}
rv_partition = tmp_vector;
//cout << "rv partition size after extra partition:" << rv_partition.size() << endl;
}
// sorting boxes by probability value
if (global_config.sort_rv_flag)
{
CLOG_IF(global_config.verbose, INFO, "algorithm")
<< "Sorting the partition of domain of continuous random parameters";
sort(rv_partition.begin(), rv_partition.end(), measure::compare_boxes_by_p_measure);
}
// getting partition of domain of discrete random variables
std::vector<box> dd_partition = measure::get_dd_partition();
// fix for now
if (dd_partition.empty())
{
dd_partition.push_back(box());
}
// generating all paths of lengths [min_depth, max_depth]
std::vector<std::vector<pdrh::mode *>> paths = pdrh::get_all_paths(min_depth, max_depth);
// initializing probability map
std::map<box, capd::interval> p_map;
capd::interval total_probability = capd::interval(0, 2 - measure::p_measure(rv_domain, global_config.precision_prob).leftBound());
for (box nd : nd_partition) {
p_map.insert(std::make_pair(nd, total_probability));
}
// initializing partition map
std::map<box, std::vector<box>> partition_map;
for (box nd : nd_partition) {
partition_map.insert(std::make_pair(nd, rv_partition));
}
vector<box> total_partition = rv_partition;
std::map<box, capd::interval> res_map, final_map;
for (auto it = p_map.begin(); it != p_map.end(); it++) {
res_map.insert(make_pair(it->first, capd::interval(0.0)));
final_map.insert(make_pair(it->first, capd::interval(0.0)));
}
// algorithm
for (box dd : dd_partition) {
// calculating the discrete measure
capd::interval dd_measure(1.0);
if (!dd.empty()) {
dd_measure = measure::p_dd_measure(dd);
}
// probability map computation starts here
while (p_map.size() > 0) {
// updating the nd_partition
nd_partition.clear();
//cout << "ND partition NOW:" << endl;
for (auto it = p_map.cbegin(); it != p_map.cend(); it++) {
nd_partition.push_back(it->first);
//cout << it->first << endl;
}
// iterating through the boxes
//for(box nd : nd_partition)
#pragma omp parallel for schedule (dynamic)
for (int j = 0; j < nd_partition.size(); j++) {
box nd = nd_partition.at(j);
#pragma omp critical
{
rv_partition = partition_map[nd];
}
vector<box> rv_stack;
//#pragma omp parallel for schedule (dynamic)
for (int i = 0; i < rv_partition.size(); i++) {
box rv = rv_partition.at(i);
// calculating probability measure of the box rv; initally p_box = [1.0, 1.0]
capd::interval p_box(1.0, 1.0);
#pragma omp critical
{
CLOG_IF(global_config.verbose, INFO, "algorithm") << "====================";
if (!nd.empty()) {
CLOG_IF(global_config.verbose, INFO, "algorithm") << "nd_box: " << nd;
}
if (!dd.empty()) {
//p_box *= measure::p_dd_measure(dd);
CLOG_IF(global_config.verbose, INFO, "algorithm") << "dd_box: " << dd;
}
if (!rv.empty()) {
p_box *= measure::p_measure(rv, global_config.precision_prob);
CLOG_IF(global_config.verbose, INFO, "algorithm") << "rv_box: " << rv;
}
CLOG_IF(global_config.verbose, INFO, "algorithm") << "p_box: " << p_box;
}
std::stringstream s;
std::string solver_opt;
#pragma omp critical
{
solver_opt = global_config.solver_opt;
s << solver_opt << " --precision ";
if (!rv.empty()) {
s << rv.min_side_width() * global_config.solver_precision_ratio;
} else {
s << global_config.solver_precision_ratio;
}
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Solver options: " << s.str();
}
switch (decision_procedure::evaluate(paths, vector<box>{nd, dd, rv}, global_config.solver_bin, s.str()))
{
case decision_procedure::result::SAT:
#pragma omp critical
{
if (p_box.leftBound() > 0) {
p_map[nd] = capd::interval(p_map[nd].leftBound() + p_box.leftBound(),
p_map[nd].rightBound());
}
CLOG_IF(global_config.verbose, INFO, "algorithm") << "SAT";
CLOG_IF(global_config.verbose, INFO, "algorithm") << "P = " << p_map[nd];
}
break;
case decision_procedure::result::UNSAT:
#pragma omp critical
{
if (p_box.leftBound() > 0) {
p_map[nd] = capd::interval(p_map[nd].leftBound(),
p_map[nd].rightBound() - p_box.leftBound());
}
CLOG_IF(global_config.verbose, INFO, "algorithm") << "UNSAT";
CLOG_IF(global_config.verbose, INFO, "algorithm") << "P = " << p_map[nd];
}
break;
case decision_procedure::result::UNDET:
#pragma omp critical
{
CLOG_IF(global_config.verbose, INFO, "algorithm") << "UNDET";
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Bisect " << rv;
std::vector<box> rv_bisect = box_factory::bisect(rv);
rv_stack.insert(rv_stack.end(), rv_bisect.begin(), rv_bisect.end());
// updating total partition
auto it = find(total_partition.begin(), total_partition.end(), rv);
if (it != total_partition.end()) {
total_partition.erase(it);
total_partition.insert(total_partition.end(), rv_bisect.begin(), rv_bisect.end());
}
}
break;
}
}
#pragma omp critical
{
if (p_map.find(nd) == p_map.end()) {
CLOG(ERROR, "algorithm") << "The box " << nd << " is not in the map";
exit(EXIT_FAILURE);
}
capd::interval probability;
probability = p_map[nd];
if (capd::intervals::width(probability) <= global_config.precision_prob) {
CLOG_IF(global_config.verbose, INFO, "algorithm")
<< "Epsilon is satisfied. Updating resulting probability map with " << nd;
p_map.erase(nd);
partition_map.erase(nd);
res_map[nd] = probability;// * dd_measure;
}
// sorting newly obtained boxes
if (global_config.sort_rv_flag) {
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Sorting bisected boxes";
sort(rv_stack.begin(), rv_stack.end(), measure::compare_boxes_by_p_measure);
}
// updating partition map only in case if probability value does not satisfy the probability precision
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Updating partition map";
if (partition_map.find(nd) != partition_map.cend()) {
partition_map[nd] = rv_stack;
}
//}
// cout << "Probability map after " << nd << " was processed" << endl;
// for(auto it2 = p_map.begin(); it2 != p_map.end(); it2++)
// {
// std::cout << it2->first << " | " << it2->second << std::endl;
// }
// cout << "Resulting map after " << nd << " was processed" << endl;
// for(auto it2 = res_map.begin(); it2 != res_map.end(); it2++)
// {
// std::cout << it2->first << " | " << it2->second << std::endl;
// }
// cout << "----------------------------------" << endl;
//exit(EXIT_FAILURE);
}
}
// cout << "----------------------------------" << endl;
// cout << "Probability map after all boxes are processed" << endl;
// for(auto it2 = p_map.begin(); it2 != p_map.end(); it2++)
// {
// std::cout << it2->first << " | " << it2->second << std::endl;
// }
// cout << "Resulting map after all boxes are processed" << endl;
// for(auto it2 = res_map.begin(); it2 != res_map.end(); it2++)
// {
// std::cout << it2->first << " | " << it2->second << std::endl;
// }
// cout << "----------------------------------" << endl;
//exit(EXIT_SUCCESS);
// updating probability and partition maps
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Updating probability map";
std::map<box, capd::interval> tmp_map = p_map;
for (auto it = tmp_map.cbegin(); it != tmp_map.cend(); it++) {
box nd = it->first;
// checking if the system features nondeterministic parameters
if (!nd.empty()) {
// bisecting the nondeterministic box
CLOG_IF(global_config.verbose, INFO, "algorithm") << "Bisect " << nd;
std::vector<box> tmp_boxes;
// checking if the --ignore-nondet flag is up
if (global_config.ignore_nondet) {
tmp_boxes = box_factory::bisect(nd);
} else {
tmp_boxes = box_factory::bisect(nd, global_config.partition_nondet_map);
}
capd::interval tmp_prob_value = p_map[nd];
std::vector<box> tmp_rv_partition = partition_map[nd];
capd::interval tmp_final_prob_value = final_map[nd];
// removing probability and partition for the bisected box
p_map.erase(nd);
partition_map.erase(nd);
res_map.erase(nd);
final_map.erase(nd);
// updating probability and partition maps
if (tmp_boxes.size() > 1) {
for (box b : tmp_boxes) {
p_map.insert(make_pair(b, tmp_prob_value));
partition_map.insert(make_pair(b, tmp_rv_partition));
res_map.insert(make_pair(b, tmp_prob_value));
final_map.insert(make_pair(b, tmp_final_prob_value));
}
} else {
res_map.insert(make_pair(nd, tmp_prob_value));
final_map.insert(make_pair(nd, tmp_final_prob_value));
}
}
}
// cout << "----------------------------------" << endl;
// cout << "Probability map after it's been updated" << endl;
// for(auto it2 = p_map.begin(); it2 != p_map.end(); it2++)
// {
// std::cout << it2->first << " | " << it2->second << std::endl;
// }
// cout << "Resulting map after it's been updated" << endl;
// for(auto it2 = res_map.begin(); it2 != res_map.end(); it2++)
// {
// std::cout << it2->first << " | " << it2->second << std::endl;
// }
// cout << "----------------------------------" << endl;
}
// probability map computation finishes here
// mutiplying every enclosure in the map by dd_measure
for (auto it = final_map.begin(); it != final_map.end(); it++) {
final_map[it->first] += res_map[it->first] * dd_measure;
}
// cout << "Final map after " << dd << endl;
// for(auto it = final_map.begin(); it != final_map.end(); it++)
// {
// cout << it->first << " | " << it->second << endl;
// }
// cout << "----------------------------------" << endl;
//return final_map;
// multiplying every enclosure by probability measure of discrete parameter
// for(auto it = res_map.begin(); it != res_map.end(); it++)
// {
// it->second *= measure::p_dd_measure(dd);
// }
// cout << "Resulting map before it's been multiplied by dd measure" << endl;
// for(auto it2 = res_map.begin(); it2 != res_map.end(); it2++)
// {
// std::cout << it2->first << " | " << it2->second << std::endl;
// }
// cout << "----------------------------------" << endl;
// updating p_map and partition_map for the next iteration
partition_map.clear();
p_map.clear();
res_map.clear();
for (auto it = final_map.begin(); it != final_map.end(); it++) {
p_map.insert(make_pair(it->first, total_probability));
res_map.insert(make_pair(it->first, capd::interval(0.0)));
partition_map.insert(make_pair(it->first, total_partition));
}
// cout << "Total partition: " << endl;
// for(box b : total_partition)
// {
// cout << b << endl;
// }
// exit(EXIT_FAILURE);
// cout << "Updated probability map after the end of the loop" << endl;
// for(auto it2 = p_map.begin(); it2 != p_map.end(); it2++)
// {
// std::cout << it2->first << " | " << it2->second << std::endl;
// }
}
// cout << "Resulting map after adding" << endl;
// for(auto it2 = res_map.begin(); it2 != res_map.end(); it2++)
// {
// cout << it2->first << " | " << it2->second << std::endl;
// }
// cout << "Final p_map" << endl;
// for(auto it2 = p_map.begin(); it2 != p_map.end(); it2++)
// {
// cout << it2->first << " | " << it2->second << std::endl;
// }
return final_map;
}