// See www.openfst.org for extensive documentation on this weighted
  // finite-state transducer library.
  //
  // Generates a random FST according to a class-specific transition model.
  
  #include <cstdlib>
  #include <ctime>
  #include <memory>
  #include <string>
  
  #include <fst/extensions/compress/randmod.h>
  
  #include <fst/flags.h>
  #include <fst/log.h>
  #include <fst/fstlib.h>
  #include <fst/util.h>
  
  DEFINE_int32(seed, time(0), "Random seed");
  DEFINE_int32(states, 10, "# of states");
  DEFINE_int32(labels, 2, "# of labels");
  DEFINE_int32(classes, 1, "# of probability distributions");
  DEFINE_bool(transducer, false, "Output a transducer");
  DEFINE_bool(weights, false, "Output a weighted FST");
  
  int main(int argc, char **argv) {
    using fst::StdVectorFst;
    using fst::StdArc;
    using fst::TropicalWeight;
    using fst::WeightGenerate;
  
    string usage = "Generates a random FST.
  
    Usage: ";
    usage += argv[0];
    usage += "[out.fst]
  ";
  
    std::set_new_handler(FailedNewHandler);
    SET_FLAGS(usage.c_str(), &argc, &argv, true);
    if (argc > 2) {
      ShowUsage();
      return 1;
    }
  
    string out_name = (argc > 1 && (strcmp(argv[1], "-") != 0)) ? argv[1] : "";
  
    srand(FLAGS_seed);
  
    int num_states = (rand() % FLAGS_states) + 1;    // NOLINT
    int num_classes = (rand() % FLAGS_classes) + 1;  // NOLINT
    int num_labels = (rand() % FLAGS_labels) + 1;    // NOLINT
  
    StdVectorFst fst;
    using TropicalWeightGenerate = WeightGenerate<TropicalWeight>;
    std::unique_ptr<TropicalWeightGenerate> generate(FLAGS_weights ?
        new TropicalWeightGenerate(false) : nullptr);
    fst::RandMod<StdArc, TropicalWeightGenerate> rand_mod(num_states,
        num_classes, num_labels, FLAGS_transducer, generate.get());
    rand_mod.Generate(&fst);
    fst.Write(out_name);
    return 0;
  }