From 3f47e8a0ce47fccc85c58b7cb3e5b50da8f41ce0 Mon Sep 17 00:00:00 2001
From: Franck Dary <franck.dary@lis-lab.fr>
Date: Sun, 17 Nov 2019 14:45:33 +0100
Subject: [PATCH] Fixed a bug in dev eval for tokenization, and removed program
options incompatible with tokenization
---
maca_common/include/ProgramParameters.hpp | 2 --
maca_common/src/ProgramParameters.cpp | 2 --
maca_common/src/programOptionsTemplates.cpp | 5 ----
trainer/src/TrainInfos.cpp | 6 -----
trainer/src/Trainer.cpp | 26 ++++-----------------
5 files changed, 4 insertions(+), 37 deletions(-)
diff --git a/maca_common/include/ProgramParameters.hpp b/maca_common/include/ProgramParameters.hpp
index c2efe89..ec8411c 100644
--- a/maca_common/include/ProgramParameters.hpp
+++ b/maca_common/include/ProgramParameters.hpp
@@ -78,8 +78,6 @@ struct ProgramParameters
static std::string tapeToMask;
static float maskRate;
static bool featureExtraction;
- static bool devEvalOnGold;
- static bool devLoss;
static bool randomDebug;
static float randomDebugProbability;
static bool alwaysSave;
diff --git a/maca_common/src/ProgramParameters.cpp b/maca_common/src/ProgramParameters.cpp
index fa4ca4c..feedefe 100644
--- a/maca_common/src/ProgramParameters.cpp
+++ b/maca_common/src/ProgramParameters.cpp
@@ -72,8 +72,6 @@ int ProgramParameters::dictCapacity;
std::string ProgramParameters::tapeToMask;
float ProgramParameters::maskRate;
bool ProgramParameters::featureExtraction;
-bool ProgramParameters::devEvalOnGold;
-bool ProgramParameters::devLoss;
bool ProgramParameters::randomDebug;
float ProgramParameters::randomDebugProbability;
bool ProgramParameters::alwaysSave;
diff --git a/maca_common/src/programOptionsTemplates.cpp b/maca_common/src/programOptionsTemplates.cpp
index d08163d..72045ae 100644
--- a/maca_common/src/programOptionsTemplates.cpp
+++ b/maca_common/src/programOptionsTemplates.cpp
@@ -102,9 +102,6 @@ po::options_description getTrainOptionsDescription()
"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");
@@ -296,7 +293,6 @@ void loadTrainProgramParameters(int argc, char * argv[])
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>();
@@ -309,7 +305,6 @@ void loadTrainProgramParameters(int argc, char * argv[])
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>();
diff --git a/trainer/src/TrainInfos.cpp b/trainer/src/TrainInfos.cpp
index ae1d6ee..ab7e106 100644
--- a/trainer/src/TrainInfos.cpp
+++ b/trainer/src/TrainInfos.cpp
@@ -300,8 +300,6 @@ void TrainInfos::setTopologyPrinted(const std::string & classifier)
{
topologyPrinted[classifier] = true;
trainLossesPerClassifierPerEpoch[classifier].emplace_back(0.0);
- if (ProgramParameters::devLoss)
- devLossesPerClassifierPerEpoch[classifier].emplace_back(0.0);
}
void TrainInfos::nextEpoch()
@@ -309,11 +307,7 @@ void TrainInfos::nextEpoch()
lastEpoch++;
saveToFilename();
for (auto & it : topologyPrinted)
- {
trainLossesPerClassifierPerEpoch[it.first].emplace_back(0.0);
- if (ProgramParameters::devLoss)
- devLossesPerClassifierPerEpoch[it.first].emplace_back(0.0);
- }
}
void TrainInfos::computeMustSaves()
diff --git a/trainer/src/Trainer.cpp b/trainer/src/Trainer.cpp
index 1c1b0b7..406c33e 100644
--- a/trainer/src/Trainer.cpp
+++ b/trainer/src/Trainer.cpp
@@ -112,19 +112,12 @@ void Trainer::computeScoreOnDev()
auto weightedActions = tm.getCurrentClassifier()->weightActions(*devConfig);
std::string pAction = "";
- std::string oAction = "";
for (auto & it : weightedActions)
if (it.first)
{
- if (pAction.empty())
- pAction = it.second.second;
-
- if (tm.getCurrentClassifier()->getActionCost(*devConfig, it.second.second) == 0)
- {
- oAction = it.second.second;
- break;
- }
+ pAction = it.second.second;
+ break;
}
if (pAction.empty())
@@ -134,18 +127,7 @@ void Trainer::computeScoreOnDev()
break;
}
- if (ProgramParameters::devLoss)
- {
- float loss = tm.getCurrentClassifier()->getLoss(*devConfig, tm.getCurrentClassifier()->getActionIndex(oAction));
- TI.addDevLoss(tm.getCurrentClassifier()->name, loss);
- }
-
- std::string actionName;
-
- if (ProgramParameters::devEvalOnGold)
- actionName = oAction;
- else
- actionName = pAction;
+ std::string actionName = pAction;
Action * action = tm.getCurrentClassifier()->getAction(actionName);
@@ -158,7 +140,7 @@ void Trainer::computeScoreOnDev()
TransitionMachine::Transition * transition = tm.getTransition(actionName);
action->setInfos(tm.getCurrentClassifier()->name);
- devConfig->addToActionsHistory(tm.getCurrentClassifier()->name, actionName, tm.getCurrentClassifier()->getActionCost(*devConfig, actionName));
+ devConfig->addToActionsHistory(tm.getCurrentClassifier()->name, actionName, 0);
action->apply(*devConfig);
tm.takeTransition(transition);
--
GitLab