From 8118cf69ca1caa62062f98688152c05c09ecb03d Mon Sep 17 00:00:00 2001
From: Franck Dary <franck.dary@lis-lab.fr>
Date: Fri, 22 Mar 2019 20:52:29 +0100
Subject: [PATCH] Added featureExtraction option
---
maca_common/include/ProgramParameters.hpp | 1 +
maca_common/src/ProgramParameters.cpp | 1 +
trainer/src/Trainer.cpp | 48 ++++++++++++++-------
trainer/src/macaon_train.cpp | 4 +-
transition_machine/include/Classifier.hpp | 2 +
transition_machine/include/FeatureModel.hpp | 4 ++
transition_machine/src/Classifier.cpp | 5 +++
transition_machine/src/FeatureModel.cpp | 14 ++++++
8 files changed, 62 insertions(+), 17 deletions(-)
diff --git a/maca_common/include/ProgramParameters.hpp b/maca_common/include/ProgramParameters.hpp
index 07f5455..274d397 100644
--- a/maca_common/include/ProgramParameters.hpp
+++ b/maca_common/include/ProgramParameters.hpp
@@ -70,6 +70,7 @@ struct ProgramParameters
static bool printOutputEntropy;
static std::string tapeToMask;
static float maskRate;
+ static bool featureExtraction;
private :
diff --git a/maca_common/src/ProgramParameters.cpp b/maca_common/src/ProgramParameters.cpp
index 4fab231..ac15fa8 100644
--- a/maca_common/src/ProgramParameters.cpp
+++ b/maca_common/src/ProgramParameters.cpp
@@ -64,4 +64,5 @@ bool ProgramParameters::printOutputEntropy;
int ProgramParameters::dictCapacity;
std::string ProgramParameters::tapeToMask;
float ProgramParameters::maskRate;
+bool ProgramParameters::featureExtraction;
diff --git a/trainer/src/Trainer.cpp b/trainer/src/Trainer.cpp
index bb8ff40..cf0c844 100644
--- a/trainer/src/Trainer.cpp
+++ b/trainer/src/Trainer.cpp
@@ -52,7 +52,7 @@ void Trainer::computeScoreOnDev()
else
{
// Print current iter advancement in percentage
- if (ProgramParameters::interactive)
+ if (ProgramParameters::interactive && !ProgramParameters::featureExtraction)
{
int totalSize = ProgramParameters::devTapeSize;
int steps = devConfig->getHead();
@@ -200,7 +200,7 @@ void Trainer::train()
}
// Print current iter advancement in percentage
- if (ProgramParameters::interactive)
+ if (ProgramParameters::interactive && !ProgramParameters::featureExtraction)
{
int totalSize = ProgramParameters::iterationSize == -1 ? ProgramParameters::tapeSize : ProgramParameters::iterationSize;
int steps = ProgramParameters::iterationSize == -1 ? trainConfig.getHead() : nbSteps;
@@ -211,27 +211,35 @@ void Trainer::train()
}
}
- auto weightedActions = tm.getCurrentClassifier()->weightActions(trainConfig);
std::string pAction = "";
std::string oAction = "";
-
bool pActionIsZeroCost = false;
- for (auto & it : weightedActions)
- if (it.first)
- {
- if (pAction == "")
- pAction = it.second.second;
+ Classifier::WeightedActions weightedActions;
+ if (!ProgramParameters::featureExtraction)
+ {
+ weightedActions = tm.getCurrentClassifier()->weightActions(trainConfig);
- if (tm.getCurrentClassifier()->getActionCost(trainConfig, it.second.second) == 0)
+ for (auto & it : weightedActions)
+ if (it.first)
{
- oAction = it.second.second;
- break;
+ if (pAction == "")
+ pAction = it.second.second;
+
+ if (tm.getCurrentClassifier()->getActionCost(trainConfig, it.second.second) == 0)
+ {
+ oAction = it.second.second;
+ break;
+ }
}
- }
- if (pAction == oAction)
- pActionIsZeroCost = true;
+ if (pAction == oAction)
+ pActionIsZeroCost = true;
+ }
+ else
+ {
+ oAction = tm.getCurrentClassifier()->getZeroCostActions(trainConfig)[0];
+ }
if (oAction.empty())
oAction = tm.getCurrentClassifier()->getDefaultAction();
@@ -252,7 +260,8 @@ void Trainer::train()
exit(1);
}
- tm.getCurrentClassifier()->trainOnExample(trainConfig, tm.getCurrentClassifier()->getActionIndex(oAction));
+ if (!ProgramParameters::featureExtraction)
+ tm.getCurrentClassifier()->trainOnExample(trainConfig, tm.getCurrentClassifier()->getActionIndex(oAction));
TI.addTrainExample(tm.getCurrentClassifier()->name);
if (pActionIsZeroCost)
@@ -262,6 +271,13 @@ void Trainer::train()
std::string actionName = "";
+ //ici
+ if (ProgramParameters::featureExtraction)
+ {
+ auto features = tm.getCurrentClassifier()->getFeatureModel()->getFeatureDescription(trainConfig).featureValues();
+ fprintf(stdout, "%s\t%s\n", oAction.c_str(), features.c_str());
+ }
+
if (TI.getEpoch() >= k && choiceWithProbability(ProgramParameters::dynamicProbability))
{
actionName = pAction;
diff --git a/trainer/src/macaon_train.cpp b/trainer/src/macaon_train.cpp
index 6062288..7885ac4 100644
--- a/trainer/src/macaon_train.cpp
+++ b/trainer/src/macaon_train.cpp
@@ -85,7 +85,8 @@ po::options_description getOptionsDescription()
"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")
+ ("printTime", "Print time on stderr.")
+ ("featureExtraction", "Use macaon only a feature extractor, print corpus to stdout.")
("shuffle", po::value<bool>()->default_value(true),
"Shuffle examples after each iteration");
@@ -268,6 +269,7 @@ int main(int argc, char * argv[])
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::trainName = vm["train"].as<std::string>();
ProgramParameters::devName = vm["dev"].as<std::string>();
ProgramParameters::lang = vm["lang"].as<std::string>();
diff --git a/transition_machine/include/Classifier.hpp b/transition_machine/include/Classifier.hpp
index 338325e..7c35d04 100644
--- a/transition_machine/include/Classifier.hpp
+++ b/transition_machine/include/Classifier.hpp
@@ -197,6 +197,8 @@ class Classifier
///
/// @return The number of actions.
unsigned int getNbActions();
+ /// @brief Get a pointer to the FeatureModel.
+ FeatureModel * getFeatureModel();
};
#endif
diff --git a/transition_machine/include/FeatureModel.hpp b/transition_machine/include/FeatureModel.hpp
index f83c8d2..052caf8 100644
--- a/transition_machine/include/FeatureModel.hpp
+++ b/transition_machine/include/FeatureModel.hpp
@@ -64,6 +64,10 @@ class FeatureModel
///
/// @return The string representing this FeatureDescription
std::string toString();
+ /// @brief Return a string representing the values of the features
+ ///
+ /// @return The string representing the values of the features
+ std::string featureValues();
};
private :
diff --git a/transition_machine/src/Classifier.cpp b/transition_machine/src/Classifier.cpp
index b5999d4..d1945e7 100644
--- a/transition_machine/src/Classifier.cpp
+++ b/transition_machine/src/Classifier.cpp
@@ -328,3 +328,8 @@ unsigned int Classifier::getNbActions()
return as->size();
}
+FeatureModel * Classifier::getFeatureModel()
+{
+ return fm.get();
+}
+
diff --git a/transition_machine/src/FeatureModel.cpp b/transition_machine/src/FeatureModel.cpp
index c65b6b4..ec6ebc3 100644
--- a/transition_machine/src/FeatureModel.cpp
+++ b/transition_machine/src/FeatureModel.cpp
@@ -133,3 +133,17 @@ FeatureModel::FeatureValue::FeatureValue()
{
}
+std::string FeatureModel::FeatureDescription::featureValues()
+{
+ std::string res;
+
+ for (auto & feature : values)
+ for (auto & value : feature.values)
+ res += value + "\t";
+
+ if (!res.empty())
+ res.pop_back();
+
+ return res;
+}
+
--
GitLab