Skip to content
Snippets Groups Projects
Select Git revision
  • 0e82071312a9c6984fb681e1a67aa4dd54cff495
  • master default protected
  • fullUD
  • movementInAction
4 results

macaon_train.cpp

Blame
    • macaon_train.cpp 14.55 KiB 
    /// @file macaon_train.cpp
/// @author Franck Dary
/// @version 1.0
/// @date 2018-08-07

#include <cstdio>
#include <cstdlib>
#include <boost/program_options.hpp>
#include "BD.hpp"
#include "Config.hpp"
#include "TransitionMachine.hpp"
#include "Trainer.hpp"
#include "ProgramParameters.hpp"

namespace po = boost::program_options;

/// @brief Get the list of mandatory and optional program arguments.
///
/// @return The lists.
po::options_description getOptionsDescription()
{
  po::options_description desc("Command-Line Arguments ");

  po::options_description req("Required");
  req.add_options()
    ("expName", po::value<std::string>()->required(),
      "Name of this experiment")
    ("templateName", po::value<std::string>()->required(),
      "Name of the template folder")
    ("tm", po::value<std::string>()->required(),
      "File describing the Tape Machine we will train")
    ("bd", po::value<std::string>()->required(),
      "BD file that describes the multi-tapes buffer")
    ("mcd", po::value<std::string>()->required(),
      "MCD file that describes the input")
    ("train,T", po::value<std::string>()->required(),
      "Training corpus formated according to the MCD");

  po::options_description opt("Optional");
  opt.add_options()
    ("help,h", "Produce this help message")
    ("debug,d", "Print infos on stderr")
    ("printEntropy", "Print mean entropy and standard deviation accross sequences")
    ("dicts", po::value<std::string>()->default_value(""),
      "The .dict file describing all the dictionaries to be used in the experiement. By default the filename specified in the .tm file will be used")
    ("featureModels", po::value<std::string>()->default_value(""),
      "For each classifier, specify what .fm (feature model) file to use. By default the filename specified in the .cla file will be used. Example : --featureModel Parser=parser.fm,Tagger=tagger.fm")
    ("optimizer", po::value<std::string>()->default_value("amsgrad"),
      "The learning algorithm to use : amsgrad | adam | sgd")
    ("loss", po::value<std::string>()->default_value("neglogsoftmax"),
      "The loss function to use : neglogsoftmax | weighted")
    ("dev", po::value<std::string>()->default_value(""),
      "Development corpus formated according to the MCD")
    ("lang", po::value<std::string>()->default_value("fr"),
      "Language you are working with")
    ("nbiter,n", po::value<int>()->default_value(5),
      "Number of training epochs (iterations)")
    ("iterationSize", po::value<int>()->default_value(-1),
      "The number of examples for each iteration. -1 means the whole training set")
    ("lr", po::value<float>()->default_value(0.001),
      "Learning rate of the optimizer")
    ("seed,s", po::value<int>()->default_value(100),
      "The random seed that will initialize RNG")
    ("nbTrain", po::value<int>()->default_value(0),
      "The number of models that will be trained, with only the random seed changing")
    ("duplicates", po::value<bool>()->default_value(true),
      "Remove identical training examples")
    ("showFeatureRepresentation", po::value<int>()->default_value(0),
      "For each state of the Config, show its feature representation")
    ("interactive", po::value<bool>()->default_value(true),
          "Is the shell interactive ? Display advancement informations")
    ("randomEmbeddings", po::value<bool>()->default_value(false),
      "When activated, the embeddings will be randomly initialized")
    ("randomParameters", po::value<bool>()->default_value(true),
      "When activated, the parameters will be randomly initialized")
    ("sequenceDelimiterTape", po::value<std::string>()->default_value("EOS"),
      "The name of the buffer's tape that contains the delimiter token for a sequence")
    ("sequenceDelimiter", po::value<std::string>()->default_value("1"),
      "The value of the token that act as a delimiter for sequences")
    ("batchSize", po::value<int>()->default_value(50),
      "The size of each minibatch (in number of taining examples)")
    ("dictCapacity", po::value<int>()->default_value(50000),
      "The maximal size of each Dict (number of differents embeddings).")
    ("tapeToMask", po::value<std::string>()->default_value("FORM"),
      "The name of the Tape for which some of the elements will be masked.")
    ("maskRate", po::value<float>()->default_value(0.0),
      "The rate of elements of the Tape that will be masked.")
    ("printTime", "Print time on stderr.")
    ("featureExtraction", "Use macaon only as a feature extractor, print corpus to stdout.")
    ("devEvalOnGold", "If true, dev accuracy will be computed on gold configurations.")
    ("devLoss", po::value<bool>()->default_value(false),
      "Compute and print total loss on dev for every epoch.")
    ("shuffle", po::value<bool>()->default_value(true),
      "Shuffle examples after each iteration");

  po::options_description oracle("Oracle related options");
  oracle.add_options()
    ("epochd", po::value<int>()->default_value(3),
      "Number of the first epoch where the oracle will be dynamic")
    ("proba", po::value<float>()->default_value(0.9),
      "The probability that the dynamic oracle will chose the predicted action");

  po::options_description ams("Amsgrad family optimizers");
  ams.add_options()
    ("b1", po::value<float>()->default_value(0.9),
      "beta1 parameter for the Amsgtad or Adam optimizer")
    ("b2", po::value<float>()->default_value(0.999),
      "beta2 parameter for the Amsgtad or Adam optimizer")
    ("bias", po::value<float>()->default_value(1e-8),
      "bias parameter for the Amsgtad or Adam  or Adagrad optimizer");

  po::options_description ga("Genetic algorithm related options");
  ga.add_options()
    ("nbIndividuals", po::value<int>()->default_value(5),
      "Number of individuals that will be used to take decisions");

  desc.add(req).add(opt).add(oracle).add(ams).add(ga);

  return desc;
}

/// @brief Store the program arguments inside a variables_map
///
/// @param od The description of all the possible options.
/// @param argc The number of arguments given to this program.
/// @param argv The values of arguments given to this program.
///
/// @return The variables map
po::variables_map checkOptions(po::options_description & od, int argc, char ** argv)
{
  po::variables_map vm;

  try {po::store(po::parse_command_line(argc, argv, od), vm);}
  catch(std::exception& e)
  {
    std::cerr << "Error: " << e.what() << "\n";
    od.print(std::cerr);
    exit(1);
  }

      if (vm.count("help"))
  {
    std::cout << od << "\n";
    exit(0);
  }

  try {po::notify(vm);}
  catch(std::exception& e)
  {
    std::cerr << "Error: " << e.what() << "\n";
    od.print(std::cerr);
    exit(1);
  }

  return vm;
}

/// @brief Set all the usefull paths relative to expPath
void updatePaths()
{
  const char * MACAON_DIR = std::getenv("MACAON_DIR");
  std::string slash = "/";
  ProgramParameters::langPath = MACAON_DIR + slash + ProgramParameters::lang + slash;
  ProgramParameters::expPath = ProgramParameters::langPath + "bin/" + ProgramParameters::expName + slash;
  ProgramParameters::templatePath = ProgramParameters::langPath + ProgramParameters::templateName + slash;
  ProgramParameters::tmFilename = ProgramParameters::expPath + ProgramParameters::tmName;
  ProgramParameters::bdFilename = ProgramParameters::expPath + ProgramParameters::bdName;
  ProgramParameters::mcdFilename = ProgramParameters::expPath + ProgramParameters::mcdName;
  ProgramParameters::trainFilename = ProgramParameters::expPath + ProgramParameters::trainName;
  ProgramParameters::devFilename = ProgramParameters::expPath + ProgramParameters::devName;
  ProgramParameters::newTemplatePath = ProgramParameters::langPath + "bin/" + ProgramParameters::baseExpName + slash;
}

/// @brief Create the folder containing the current experiment from the template frolder
void createExpPath()
{
  // If this is the case, the training should resume where it was stopped.
  // No need to recreate the expPath.
  if (ProgramParameters::newTemplatePath == ProgramParameters::expPath)
    return;

std::string decode = "\
#! /bin/bash\n\
\n\
if [ \"$#\" -lt 2 ]; then\n\
 echo \"Usage : $0 input mcd\"\n\
 exit\n\
fi\n\
\n\
INPUT=$1\n\
MCD=$2\n\
\n\
shift\n\
shift\n\
ARGS=\"\"\n\
for arg in \"$@\"\n\
do\n\
  ARGS=\"$ARGS $arg\"\n\
done\n\
\n\
macaon_decode --lang " + ProgramParameters::lang +  " --tm machine.tm --bd test.bd -I $INPUT --mcd $MCD --expName " + ProgramParameters::expName + "$ARGS";

  if (system(("rm -r " + ProgramParameters::expPath + " 2> /dev/null").c_str())){}
  if (system(("mkdir " + ProgramParameters::expPath).c_str())){}
  if (system(("cp -r " + ProgramParameters::newTemplatePath + "* " + ProgramParameters::expPath + ".").c_str())){}
  if (system(("echo \'" + decode + "\' > " + ProgramParameters::expPath + "decode.sh").c_str())){}
  if (system(("chmod +x " + ProgramParameters::expPath + "decode.sh").c_str())){}
  if (system(("ln -f -s " + ProgramParameters::expPath + "decode.sh " + ProgramParameters::langPath + "bin/maca_tm_" + ProgramParameters::expName).c_str())){}

  ProgramParameters::tapeSize = getNbLines(ProgramParameters::trainFilename);
      ProgramParameters::devTapeSize = ProgramParameters::devFilename.empty() ? 0 : getNbLines(ProgramParameters::devFilename);
  ProgramParameters::readSize = ProgramParameters::tapeSize;
}

/// @brief Train a model according to all the ProgramParameters
void launchTraining()
{
  TransitionMachine transitionMachine(true);

  BD trainBD(ProgramParameters::bdFilename, ProgramParameters::mcdFilename);
  Config trainConfig(trainBD, ProgramParameters::trainFilename);

  std::unique_ptr<BD> devBD;
  std::unique_ptr<Config> devConfig;

  std::unique_ptr<Trainer> trainer;

  if(ProgramParameters::devFilename.empty())
  {
    trainer.reset(new Trainer(transitionMachine, trainBD, trainConfig));
  }
  else
  {
    devBD.reset(new BD(ProgramParameters::bdFilename, ProgramParameters::mcdFilename));
    devConfig.reset(new Config(*devBD.get(), ProgramParameters::devFilename));
    trainer.reset(new Trainer(transitionMachine, trainBD, trainConfig, devBD.get(), devConfig.get()));
  }

  trainer->train();
}

void createTemplatePath()
{
  if (system(("rm -r " + ProgramParameters::newTemplatePath + " 2> /dev/null").c_str())){}
  if (system(("mkdir " + ProgramParameters::newTemplatePath).c_str())){}
  if (system(("cp -r " + ProgramParameters::templatePath + "* " + ProgramParameters::newTemplatePath + ".").c_str())){}
}

void removeTemplatePath()
{
  if (system(("rm -r " + ProgramParameters::newTemplatePath + " 2> /dev/null").c_str())){}
}

/// @brief Train a TransitionMachine to predict and add information to a structured input file, by using annotated examples.
///
/// @param argc The number of arguments given to this program.
/// @param argv[] Array of arguments given to this program.
///
/// @return 0 if there was no crash.
int main(int argc, char * argv[])
{
  auto od = getOptionsDescription();

  po::variables_map vm = checkOptions(od, argc, argv);

  ProgramParameters::expName = vm["expName"].as<std::string>();
  ProgramParameters::baseExpName = ProgramParameters::expName;
  ProgramParameters::templateName = vm["templateName"].as<std::string>();
  ProgramParameters::tmName = vm["tm"].as<std::string>();
  ProgramParameters::bdName = vm["bd"].as<std::string>();
  ProgramParameters::mcdName = vm["mcd"].as<std::string>();
  ProgramParameters::debug = vm.count("debug") == 0 ? false : true;
  ProgramParameters::printEntropy = vm.count("printEntropy") == 0 ? false : true;
  ProgramParameters::printTime = vm.count("printTime") == 0 ? false : true;
  ProgramParameters::featureExtraction = vm.count("featureExtraction") == 0 ? false : true;
  ProgramParameters::devEvalOnGold = vm.count("devEvalOnGold") == 0 ? false : true;
  ProgramParameters::trainName = vm["train"].as<std::string>();
  ProgramParameters::devName = vm["dev"].as<std::string>();
  ProgramParameters::lang = vm["lang"].as<std::string>();
  ProgramParameters::nbIter = vm["nbiter"].as<int>();
      ProgramParameters::seed = vm["seed"].as<int>();
  ProgramParameters::batchSize = vm["batchSize"].as<int>();
  ProgramParameters::dictCapacity = vm["dictCapacity"].as<int>();
  ProgramParameters::nbTrain = vm["nbTrain"].as<int>();
  ProgramParameters::removeDuplicates = vm["duplicates"].as<bool>();
  ProgramParameters::interactive = vm["interactive"].as<bool>();
  ProgramParameters::shuffleExamples = vm["shuffle"].as<bool>();
  ProgramParameters::devLoss = vm["devLoss"].as<bool>();
  ProgramParameters::randomEmbeddings = vm["randomEmbeddings"].as<bool>();
  ProgramParameters::randomParameters = vm["randomParameters"].as<bool>();
  ProgramParameters::sequenceDelimiterTape = vm["sequenceDelimiterTape"].as<std::string>();
  ProgramParameters::sequenceDelimiter = vm["sequenceDelimiter"].as<std::string>();
  ProgramParameters::learningRate = vm["lr"].as<float>();
  ProgramParameters::beta1 = vm["b1"].as<float>();
  ProgramParameters::beta2 = vm["b2"].as<float>();
  ProgramParameters::bias = vm["bias"].as<float>();
  ProgramParameters::nbIndividuals = vm["nbIndividuals"].as<int>();
  ProgramParameters::optimizer = vm["optimizer"].as<std::string>();
  ProgramParameters::dicts = vm["dicts"].as<std::string>();
  ProgramParameters::loss = vm["loss"].as<std::string>();
  ProgramParameters::dynamicEpoch = vm["epochd"].as<int>();
  ProgramParameters::dynamicProbability = vm["proba"].as<float>();
  ProgramParameters::tapeToMask = vm["tapeToMask"].as<std::string>();
  ProgramParameters::maskRate = vm["maskRate"].as<float>();
  ProgramParameters::showFeatureRepresentation = vm["showFeatureRepresentation"].as<int>();
  ProgramParameters::iterationSize = vm["iterationSize"].as<int>();
  std::string featureModels = vm["featureModels"].as<std::string>();
  if (!featureModels.empty())
  {
    auto byClassifiers = split(featureModels, ',');
    for (auto & classifier : byClassifiers)
    {
      auto parts = split(classifier, '=');
      if (parts.size() != 2)
      {
        fprintf(stderr, "ERROR (%s) : wrong format for argument of option featureModels. Aborting.\n", ERRINFO);
        exit(1);
      }
      ProgramParameters::featureModelByClassifier[parts[0]] = parts[1];
    }
  }

  if (ProgramParameters::nbTrain)
  {
    updatePaths();
    createTemplatePath();
    for (int i = 0; i < ProgramParameters::nbTrain; i++)
    {
      fprintf(stderr, "Training number %d / %d :\n", i+1, ProgramParameters::nbTrain);
      ProgramParameters::expName = ProgramParameters::baseExpName + "_" + std::to_string(i);
      updatePaths();
      createExpPath();
      Dict::deleteDicts();
      launchTraining();
    }
    removeTemplatePath();
  }
  else
  {
    updatePaths();
    ProgramParameters::newTemplatePath = ProgramParameters::templatePath;
    createExpPath();
    launchTraining();
  }

  return 0;
}