diff --git a/decoder/src/Decoder.cpp b/decoder/src/Decoder.cpp
index 273e9a555d350433e8adbf7e3196ae0505f15acb..94dd16beb55786200195827d18f708e0202a188a 100644
--- a/decoder/src/Decoder.cpp
+++ b/decoder/src/Decoder.cpp
@@ -5,79 +5,134 @@ Decoder::Decoder(ReadingMachine & machine) : machine(machine)
{
}
-void Decoder::decode(BaseConfig & config, std::size_t beamSize, bool debug, bool printAdvancement)
+void Decoder::decode(BaseConfig & baseConfig, std::size_t beamSize, bool debug, bool printAdvancement)
{
torch::AutoGradMode useGrad(false);
machine.trainMode(false);
machine.setDictsState(Dict::State::Closed);
machine.getStrategy().reset();
- config.addPredicted(machine.getPredicted());
constexpr int printInterval = 50;
int nbExamplesProcessed = 0;
auto pastTime = std::chrono::high_resolution_clock::now();
+ std::vector<BaseConfig> beam;
+ std::vector<bool> endFlag;
+
try
{
- config.setState(machine.getStrategy().getInitialState());
+
+ baseConfig.addPredicted(machine.getPredicted());
+ baseConfig.setState(machine.getStrategy().getInitialState());
+ for (unsigned int i = 0; i < beamSize; i++)
+ {
+ beam.emplace_back(baseConfig);
+ endFlag.emplace_back(false);
+ }
+
machine.getClassifier()->setState(machine.getStrategy().getInitialState());
while (true)
{
- if (debug)
- config.printForDebug(stderr);
-
if (machine.hasSplitWordTransitionSet())
- config.setAppliableSplitTransitions(machine.getSplitWordTransitionSet().getNAppliableTransitions(config, Config::maxNbAppliableSplitTransitions));
- auto appliableTransitions = machine.getTransitionSet().getAppliableTransitions(config);
- config.setAppliableTransitions(appliableTransitions);
+ for (auto & c : beam)
+ c.setAppliableSplitTransitions(machine.getSplitWordTransitionSet().getNAppliableTransitions(c, Config::maxNbAppliableSplitTransitions));
- auto context = machine.getClassifier()->getNN()->extractContext(config).back();
-
- auto neuralInput = torch::from_blob(context.data(), {(long)context.size()}, torch::kLong).clone().to(NeuralNetworkImpl::device);
- auto prediction = machine.getClassifier()->getNN()(neuralInput).squeeze();
+ std::vector<std::vector<int>> appliableTransitions;
+ for (auto & c : beam)
+ {
+ appliableTransitions.emplace_back(machine.getTransitionSet().getAppliableTransitions(c));
+ c.setAppliableTransitions(appliableTransitions.back());
+ }
- int chosenTransition = -1;
- float bestScore = std::numeric_limits<float>::min();
+ std::vector<torch::Tensor> predictions;
+ for (auto & c : beam)
+ {
+ machine.getClassifier()->setState(c.getState());
+ auto context = machine.getClassifier()->getNN()->extractContext(c).back();
+ auto neuralInput = torch::from_blob(context.data(), {(long)context.size()}, torch::kLong).clone().to(NeuralNetworkImpl::device);
+ predictions.emplace_back(machine.getClassifier()->getNN()(neuralInput).squeeze());
+ }
if (debug)
{
- auto softmaxed = torch::softmax(prediction,-1);
- std::vector<std::pair<float,std::string>> toPrint;
- for (unsigned int i = 0; i < softmaxed.size(0); i++)
+ fmt::print(stderr, "{:-<{}}\n", "", 80);
+ fmt::print(stderr, "BEAM SEARCH CONTENT :\n");
+ for (unsigned int beamIndex = 0; beamIndex < beam.size(); beamIndex++)
{
- float score = softmaxed[i].item<float>();
- std::string nicePrint = fmt::format("{} {:7.2f} {}", appliableTransitions[i] ? "*" : " ", score, machine.getTransitionSet().getTransition(i)->getName());
- toPrint.emplace_back(std::make_pair(score,nicePrint));
+ auto & c = beam[beamIndex];
+ c.printForDebug(stderr);
+ auto softmaxed = torch::softmax(predictions[beamIndex],-1);
+ std::vector<std::pair<float,std::string>> toPrint;
+ for (unsigned int i = 0; i < softmaxed.size(0); i++)
+ {
+ float score = softmaxed[i].item<float>();
+ std::string nicePrint = fmt::format("{} {:7.2f} {}", appliableTransitions[beamIndex][i] ? "*" : " ", score, machine.getTransitionSet().getTransition(i)->getName());
+ toPrint.emplace_back(std::make_pair(score,nicePrint));
+ }
+ std::sort(toPrint.rbegin(), toPrint.rend());
+ for (unsigned int i = 0; i < 5 and i < toPrint.size(); i++)
+ fmt::print(stderr, "{}\n", toPrint[i].second);
}
- std::sort(toPrint.rbegin(), toPrint.rend());
- for (unsigned int i = 0; i < 5 and i < toPrint.size(); i++)
- fmt::print(stderr, "{}\n", toPrint[i].second);
+ fmt::print(stderr, "END OF BEAM SEARCH CONTENT\n");
+ fmt::print(stderr, "{:-<{}}\n", "", 80);
}
- try
+ for (unsigned int beamIndex = 0; beamIndex < beam.size(); beamIndex++)
{
- for (unsigned int i = 0; i < prediction.size(0); i++)
+ if (endFlag[beamIndex])
+ continue;
+ auto & c = beam[beamIndex];
+ int chosenTransition = -1;
+ float bestScore = std::numeric_limits<float>::min();
+
+ try
{
- float score = prediction[i].item<float>();
- if ((chosenTransition == -1 or score > bestScore) and appliableTransitions[i])
+ for (unsigned int i = 0; i < predictions[beamIndex].size(0); i++)
{
- chosenTransition = i;
- bestScore = score;
+ float score = predictions[beamIndex][i].item<float>();
+ if ((chosenTransition == -1 or score > bestScore) and appliableTransitions[beamIndex][i])
+ {
+ chosenTransition = i;
+ bestScore = score;
+ }
}
+ } catch(std::exception & e) {util::myThrow(e.what());}
+
+ if (chosenTransition == -1)
+ {
+ c.printForDebug(stderr);
+ util::myThrow("No transition appliable !");
}
- } catch(std::exception & e) {util::myThrow(e.what());}
- if (chosenTransition == -1)
- {
- config.printForDebug(stderr);
- util::myThrow("No transition appliable !");
+ auto * transition = machine.getTransitionSet().getTransition(chosenTransition);
+
+ transition->apply(c);
+ c.addToHistory(transition->getName());
+
+ auto movement = machine.getStrategy().getMovement(c, transition->getName());
+ if (debug)
+ {
+ //TODO improve this for beam search
+ fmt::print(stderr, "(Transition,Newstate,Movement) = ({},{},{})\n", transition->getName(), movement.first, movement.second);
+ }
+ if (movement == Strategy::endMovement)
+ {
+ endFlag[beamIndex] = true;
+ continue;
+ }
+
+ c.setState(movement.first);
+ c.moveWordIndexRelaxed(movement.second);
}
- auto * transition = machine.getTransitionSet().getTransition(chosenTransition);
+ bool allBeamAreEnded = true;
+ for (unsigned int i = 0; i < beam.size(); i++)
+ if (!endFlag[i])
+ allBeamAreEnded = false;
- transition->apply(config);
- config.addToHistory(transition->getName());
+ if (allBeamAreEnded)
+ break;
if (printAdvancement)
if (++nbExamplesProcessed >= printInterval)
@@ -89,32 +144,29 @@ void Decoder::decode(BaseConfig & config, std::size_t beamSize, bool debug, bool
nbExamplesProcessed = 0;
}
- auto movement = machine.getStrategy().getMovement(config, transition->getName());
- if (debug)
- fmt::print(stderr, "(Transition,Newstate,Movement) = ({},{},{})\n", transition->getName(), movement.first, movement.second);
- if (movement == Strategy::endMovement)
- break;
-
- config.setState(movement.first);
- machine.getClassifier()->setState(movement.first);
- config.moveWordIndexRelaxed(movement.second);
}
+
} catch(std::exception & e) {util::myThrow(e.what());}
- // Force EOS when needed
- if (machine.getTransitionSet().getTransition("EOS b.0") and config.getLastNotEmptyHypConst(Config::EOSColName, config.getWordIndex()) != Config::EOSSymbol1)
+ for (auto & c : beam)
{
- machine.getTransitionSet().getTransition("EOS b.0")->apply(config);
- if (debug)
+ // Force EOS when needed
+ if (machine.getTransitionSet().getTransition("EOS b.0") and c.getLastNotEmptyHypConst(Config::EOSColName, c.getWordIndex()) != Config::EOSSymbol1)
{
- fmt::print(stderr, "Forcing EOS transition\n");
- config.printForDebug(stderr);
+ machine.getTransitionSet().getTransition("EOS b.0")->apply(c);
+ if (debug)
+ {
+ fmt::print(stderr, "Forcing EOS transition\n");
+ c.printForDebug(stderr);
+ }
}
+
+ // Fill holes in important columns like "ID" and "HEAD" to be compatible with eval script
+ try {c.addMissingColumns();}
+ catch (std::exception & e) {util::myThrow(e.what());}
}
- // Fill holes in important columns like "ID" and "HEAD" to be compatible with eval script
- try {config.addMissingColumns();}
- catch (std::exception & e) {util::myThrow(e.what());}
+ baseConfig = beam[0];
}
float Decoder::getMetricScore(const std::string & metric, std::size_t scoreIndex) const
diff --git a/decoder/src/MacaonDecode.cpp b/decoder/src/MacaonDecode.cpp
index 8aac97762dac362506ccd3ecc1e4330f5a9e1903..290f47f582b2a1031ce0ef9b6d8a0108ce9d4ee6 100644
--- a/decoder/src/MacaonDecode.cpp
+++ b/decoder/src/MacaonDecode.cpp
@@ -22,6 +22,8 @@ po::options_description MacaonDecode::getOptionsDescription()
opt.add_options()
("debug,d", "Print debuging infos on stderr")
("silent", "Don't print speed and progress")
+ ("beamSize", po::value<int>()->default_value(1),
+ "Size of the beam during beam search")
("help,h", "Produce this help message");
desc.add(req).add(opt);
@@ -71,6 +73,7 @@ int MacaonDecode::main()
auto mcdFile = variables["mcd"].as<std::string>();
bool debug = variables.count("debug") == 0 ? false : true;
bool printAdvancement = !debug && variables.count("silent") == 0 ? true : false;
+ auto beamSize = variables["beamSize"].as<int>();
torch::globalContext().setBenchmarkCuDNN(true);
@@ -86,7 +89,7 @@ int MacaonDecode::main()
BaseConfig config(mcdFile, inputTSV, inputTXT);
- decoder.decode(config, 1, debug, printAdvancement);
+ decoder.decode(config, beamSize, debug, printAdvancement);
config.print(stdout);
} catch(std::exception & e) {util::error(e);}
diff --git a/reading_machine/include/Config.hpp b/reading_machine/include/Config.hpp
index 527fd23fefcb25387f9396e7d8be5e89731fef0e..b9de6887e894852d3716d994fa9e63e3e548e71f 100644
--- a/reading_machine/include/Config.hpp
+++ b/reading_machine/include/Config.hpp
@@ -52,7 +52,7 @@ class Config
protected :
- const Utf8String & rawInput;
+ const Utf8String * rawInput;
std::size_t wordIndex{0};
std::size_t characterIndex{0};
String state{"NONE"};
diff --git a/reading_machine/src/Config.cpp b/reading_machine/src/Config.cpp
index 320686c1564b25c82176ddfde95dfb41e8495c91..4a218a5daefc7eed4398c67f5ec14429896c1f3c 100644
--- a/reading_machine/src/Config.cpp
+++ b/reading_machine/src/Config.cpp
@@ -1,7 +1,7 @@
#include "Config.hpp"
#include "util.hpp"
-Config::Config(const Utf8String & rawInput) : rawInput(rawInput)
+Config::Config(const Utf8String & rawInput) : rawInput(&rawInput)
{
}
@@ -208,9 +208,9 @@ void Config::printForDebug(FILE * dest) const
if (!stackStr.empty())
stackStr.pop_back();
fmt::print(dest, "{}\n", longLine);
- for (std::size_t index = characterIndex; index < util::getSize(rawInput) and index - characterIndex < lettersWindowSize; index++)
+ for (std::size_t index = characterIndex; index < util::getSize(*rawInput) and index - characterIndex < lettersWindowSize; index++)
fmt::print(dest, "{}", getLetter(index));
- if (rawInput.size())
+ if (rawInput->size())
fmt::print(dest, "\n{}\n", longLine);
fmt::print(dest, "State={}\nwordIndex={} characterIndex={}\nhistory=({})\nstack=({})\n", state, wordIndex, characterIndex, historyStr, stackStr);
fmt::print(dest, "{}\n", longLine);
@@ -366,12 +366,12 @@ void Config::swapStack(int relIndex1, int relIndex2)
bool Config::hasCharacter(int letterIndex) const
{
- return letterIndex >= 0 and letterIndex < (int)util::getSize(rawInput);
+ return letterIndex >= 0 and letterIndex < (int)util::getSize(*rawInput);
}
util::utf8char Config::getLetter(int letterIndex) const
{
- return rawInput[letterIndex];
+ return (*rawInput)[letterIndex];
}
bool Config::isComment(std::size_t lineIndex) const
@@ -495,20 +495,20 @@ bool Config::canMoveWordIndex(int relativeMovement) const
bool Config::moveCharacterIndex(int relativeMovement)
{
int oldVal = characterIndex;
- characterIndex = std::max(0, (int)std::min(characterIndex+relativeMovement, util::getSize(rawInput)));
+ characterIndex = std::max(0, (int)std::min(characterIndex+relativeMovement, util::getSize(*rawInput)));
return (int)characterIndex == oldVal + relativeMovement;
}
bool Config::canMoveCharacterIndex(int relativeMovement) const
{
- int target = std::max(0, (int)std::min(characterIndex+relativeMovement, util::getSize(rawInput)));
+ int target = std::max(0, (int)std::min(characterIndex+relativeMovement, util::getSize(*rawInput)));
return target == (int)characterIndex + relativeMovement;
}
bool Config::rawInputOnlySeparatorsLeft() const
{
- for (unsigned int i = characterIndex; i < rawInput.size(); i++)
- if (!util::isSeparator(rawInput[i]))
+ for (unsigned int i = characterIndex; i < rawInput->size(); i++)
+ if (!util::isSeparator((*rawInput)[i]))
return false;
return true;
@@ -565,7 +565,7 @@ void Config::setState(const std::string state)
bool Config::stateIsDone() const
{
- if (!rawInput.empty())
+ if (!rawInput->empty())
return rawInputOnlySeparatorsLeft() and !has(0, wordIndex+1, 0) and !hasStack(0);
return !has(0, wordIndex+1, 0) and !hasStack(0);
diff --git a/reading_machine/src/SubConfig.cpp b/reading_machine/src/SubConfig.cpp
index c571b3c95bdbf2efeb4fa1647c168acef99b49a2..161fc635e077584f3a88da43b2f533b530006d04 100644
--- a/reading_machine/src/SubConfig.cpp
+++ b/reading_machine/src/SubConfig.cpp
@@ -1,6 +1,6 @@
#include "SubConfig.hpp"
-SubConfig::SubConfig(BaseConfig & model, std::size_t spanSize) : Config(model.rawInput), model(model), spanSize(spanSize)
+SubConfig::SubConfig(BaseConfig & model, std::size_t spanSize) : Config(*model.rawInput), model(model), spanSize(spanSize)
{
wordIndex = model.wordIndex;
characterIndex = model.characterIndex;