/usr/include/fst/shortest-distance.h is in libfst-dev 1.5.3+r3-2.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 | // See www.openfst.org for extensive documentation on this weighted
// finite-state transducer library.
//
// Functions and classes to find shortest distance in an FST.
#ifndef FST_LIB_SHORTEST_DISTANCE_H_
#define FST_LIB_SHORTEST_DISTANCE_H_
#include <deque>
#include <vector>
#include <fst/arcfilter.h>
#include <fst/cache.h>
#include <fst/queue.h>
#include <fst/reverse.h>
#include <fst/test-properties.h>
namespace fst {
template <class Arc, class Queue, class ArcFilter>
struct ShortestDistanceOptions {
typedef typename Arc::StateId StateId;
Queue *state_queue; // Queue discipline used; owned by caller
ArcFilter arc_filter; // Arc filter (e.g., limit to only epsilon graph)
StateId source; // If kNoStateId, use the Fst's initial state
float delta; // Determines the degree of convergence required
bool first_path; // For a semiring with the path property (o.w.
// undefined), compute the shortest-distances along
// along the first path to a final state found
// by the algorithm. That path is the shortest-path
// only if the FST has a unique final state (or all
// the final states have the same final weight), the
// queue discipline is shortest-first and all the
// weights in the FST are between One() and Zero()
// according to NaturalLess.
ShortestDistanceOptions(Queue *q, ArcFilter filt, StateId src = kNoStateId,
float d = kDelta)
: state_queue(q),
arc_filter(filt),
source(src),
delta(d),
first_path(false) {}
};
// Computation state of the shortest-distance algorithm. Reusable
// information is maintained across calls to member function
// ShortestDistance(source) when 'retain' is true for improved
// efficiency when calling multiple times from different source states
// (e.g., in epsilon removal). Contrary to usual conventions, 'fst'
// may not be freed before this class. Vector 'distance' should not be
// modified by the user between these calls.
// The Error() method returns true if an error was encountered.
template <class Arc, class Queue, class ArcFilter>
class ShortestDistanceState {
public:
typedef typename Arc::StateId StateId;
typedef typename Arc::Weight Weight;
ShortestDistanceState(
const Fst<Arc> &fst, std::vector<Weight> *distance,
const ShortestDistanceOptions<Arc, Queue, ArcFilter> &opts, bool retain)
: fst_(fst),
distance_(distance),
state_queue_(opts.state_queue),
arc_filter_(opts.arc_filter),
delta_(opts.delta),
first_path_(opts.first_path),
retain_(retain),
source_id_(0),
error_(false) {
distance_->clear();
}
~ShortestDistanceState() {}
void ShortestDistance(StateId source);
bool Error() const { return error_; }
private:
const Fst<Arc> &fst_;
std::vector<Weight> *distance_;
Queue *state_queue_;
ArcFilter arc_filter_;
float delta_;
bool first_path_;
bool retain_; // Retain and reuse information across calls
std::vector<Weight> rdistance_; // Relaxation distance.
std::vector<bool> enqueued_; // Is state enqueued?
std::vector<StateId> sources_; // Source ID for ith state in 'distance_',
// 'rdistance_', and 'enqueued_' if retained.
StateId source_id_; // Unique ID characterizing each call to SD
bool error_;
};
// Compute the shortest distance. If 'source' is kNoStateId, use
// the initial state of the Fst.
template <class Arc, class Queue, class ArcFilter>
void ShortestDistanceState<Arc, Queue, ArcFilter>::ShortestDistance(
StateId source) {
if (fst_.Start() == kNoStateId) {
if (fst_.Properties(kError, false)) error_ = true;
return;
}
if (!(Weight::Properties() & kRightSemiring)) {
FSTERROR() << "ShortestDistance: Weight needs to be right distributive: "
<< Weight::Type();
error_ = true;
return;
}
if (first_path_ && !(Weight::Properties() & kPath)) {
FSTERROR() << "ShortestDistance: The first_path option is disallowed when "
<< "Weight does not have the path property: " << Weight::Type();
error_ = true;
return;
}
state_queue_->Clear();
if (!retain_) {
distance_->clear();
rdistance_.clear();
enqueued_.clear();
}
if (source == kNoStateId) source = fst_.Start();
while (distance_->size() <= source) {
distance_->push_back(Weight::Zero());
rdistance_.push_back(Weight::Zero());
enqueued_.push_back(false);
}
if (retain_) {
while (sources_.size() <= source) sources_.push_back(kNoStateId);
sources_[source] = source_id_;
}
(*distance_)[source] = Weight::One();
rdistance_[source] = Weight::One();
enqueued_[source] = true;
state_queue_->Enqueue(source);
while (!state_queue_->Empty()) {
StateId s = state_queue_->Head();
state_queue_->Dequeue();
while (distance_->size() <= s) {
distance_->push_back(Weight::Zero());
rdistance_.push_back(Weight::Zero());
enqueued_.push_back(false);
}
if (first_path_ && (fst_.Final(s) != Weight::Zero())) break;
enqueued_[s] = false;
Weight r = rdistance_[s];
rdistance_[s] = Weight::Zero();
for (ArcIterator<Fst<Arc>> aiter(fst_, s); !aiter.Done(); aiter.Next()) {
const Arc &arc = aiter.Value();
if (!arc_filter_(arc)) continue;
while (distance_->size() <= arc.nextstate) {
distance_->push_back(Weight::Zero());
rdistance_.push_back(Weight::Zero());
enqueued_.push_back(false);
}
if (retain_) {
while (sources_.size() <= arc.nextstate) sources_.push_back(kNoStateId);
if (sources_[arc.nextstate] != source_id_) {
(*distance_)[arc.nextstate] = Weight::Zero();
rdistance_[arc.nextstate] = Weight::Zero();
enqueued_[arc.nextstate] = false;
sources_[arc.nextstate] = source_id_;
}
}
Weight &nd = (*distance_)[arc.nextstate];
Weight &nr = rdistance_[arc.nextstate];
Weight w = Times(r, arc.weight);
if (!ApproxEqual(nd, Plus(nd, w), delta_)) {
nd = Plus(nd, w);
nr = Plus(nr, w);
if (!nd.Member() || !nr.Member()) {
error_ = true;
return;
}
if (!enqueued_[arc.nextstate]) {
state_queue_->Enqueue(arc.nextstate);
enqueued_[arc.nextstate] = true;
} else {
state_queue_->Update(arc.nextstate);
}
}
}
}
++source_id_;
if (fst_.Properties(kError, false)) error_ = true;
}
// Shortest-distance algorithm: this version allows fine control
// via the options argument. See below for a simpler interface.
//
// This computes the shortest distance from the 'opts.source' state to
// each visited state S and stores the value in the 'distance' vector.
// An unvisited state S has distance Zero(), which will be stored in
// the 'distance' vector if S is less than the maximum visited state.
// The state queue discipline, arc filter, and convergence delta are
// taken in the options argument.
// The 'distance' vector will contain a unique element for which
// Member() is false if an error was encountered.
//
// The weights must must be right distributive and k-closed (i.e., 1 +
// x + x^2 + ... + x^(k +1) = 1 + x + x^2 + ... + x^k).
//
// The algorithm is from Mohri, "Semiring Framework and Algorithms for
// Shortest-Distance Problems", Journal of Automata, Languages and
// Combinatorics 7(3):321-350, 2002. The complexity of algorithm
// depends on the properties of the semiring and the queue discipline
// used. Refer to the paper for more details.
template <class Arc, class Queue, class ArcFilter>
void ShortestDistance(
const Fst<Arc> &fst, std::vector<typename Arc::Weight> *distance,
const ShortestDistanceOptions<Arc, Queue, ArcFilter> &opts) {
ShortestDistanceState<Arc, Queue, ArcFilter> sd_state(fst, distance, opts,
false);
sd_state.ShortestDistance(opts.source);
if (sd_state.Error()) {
distance->clear();
distance->resize(1, Arc::Weight::NoWeight());
}
}
// Shortest-distance algorithm: simplified interface. See above for a
// version that allows finer control.
//
// If 'reverse' is false, this computes the shortest distance from the
// initial state to each state S and stores the value in the
// 'distance' vector. If 'reverse' is true, this computes the shortest
// distance from each state to the final states. An unvisited state S
// has distance Zero(), which will be stored in the 'distance' vector
// if S is less than the maximum visited state. The state queue
// discipline is automatically-selected.
// The 'distance' vector will contain a unique element for which
// Member() is false if an error was encountered.
//
// The weights must must be right (left) distributive if reverse is
// false (true) and k-closed (i.e., 1 + x + x^2 + ... + x^(k +1) = 1 +
// x + x^2 + ... + x^k).
//
// Arc weights must satisfy the property that the sum of the weights of one or
// more paths from some state S to T is never Zero(). In particular, arc weights
// are never Zero().
//
// The algorithm is from Mohri, "Semiring Framweork and Algorithms for
// Shortest-Distance Problems", Journal of Automata, Languages and
// Combinatorics 7(3):321-350, 2002. The complexity of algorithm
// depends on the properties of the semiring and the queue discipline
// used. Refer to the paper for more details.
template <class Arc>
void ShortestDistance(const Fst<Arc> &fst,
std::vector<typename Arc::Weight> *distance,
bool reverse = false, float delta = kDelta) {
typedef typename Arc::StateId StateId;
typedef typename Arc::Weight Weight;
if (!reverse) {
AnyArcFilter<Arc> arc_filter;
AutoQueue<StateId> state_queue(fst, distance, arc_filter);
ShortestDistanceOptions<Arc, AutoQueue<StateId>, AnyArcFilter<Arc>> opts(
&state_queue, arc_filter);
opts.delta = delta;
ShortestDistance(fst, distance, opts);
} else {
typedef ReverseArc<Arc> ReverseArc;
typedef typename ReverseArc::Weight ReverseWeight;
AnyArcFilter<ReverseArc> rarc_filter;
VectorFst<ReverseArc> rfst;
Reverse(fst, &rfst);
std::vector<ReverseWeight> rdistance;
AutoQueue<StateId> state_queue(rfst, &rdistance, rarc_filter);
ShortestDistanceOptions<ReverseArc, AutoQueue<StateId>,
AnyArcFilter<ReverseArc>> ropts(&state_queue,
rarc_filter);
ropts.delta = delta;
ShortestDistance(rfst, &rdistance, ropts);
distance->clear();
if (rdistance.size() == 1 && !rdistance[0].Member()) {
distance->resize(1, Arc::Weight::NoWeight());
return;
}
while (distance->size() < rdistance.size() - 1)
distance->push_back(rdistance[distance->size() + 1].Reverse());
}
}
// Return the sum of the weight of all successful paths in an FST, i.e.,
// the shortest-distance from the initial state to the final states.
// Returns a weight such that Member() is false if an error was encountered.
template <class Arc>
typename Arc::Weight ShortestDistance(const Fst<Arc> &fst,
float delta = kDelta) {
typedef typename Arc::Weight Weight;
typedef typename Arc::StateId StateId;
std::vector<Weight> distance;
if (Weight::Properties() & kRightSemiring) {
ShortestDistance(fst, &distance, false, delta);
if (distance.size() == 1 && !distance[0].Member())
return Arc::Weight::NoWeight();
Weight sum = Weight::Zero();
for (StateId s = 0; s < distance.size(); ++s)
sum = Plus(sum, Times(distance[s], fst.Final(s)));
return sum;
} else {
ShortestDistance(fst, &distance, true, delta);
StateId s = fst.Start();
if (distance.size() == 1 && !distance[0].Member())
return Arc::Weight::NoWeight();
return s != kNoStateId && s < distance.size() ? distance[s]
: Weight::Zero();
}
}
} // namespace fst
#endif // FST_LIB_SHORTEST_DISTANCE_H_
|