From 0c86cb53315b168fc479310660cce1b8d7072b5b Mon Sep 17 00:00:00 2001
From: Franck Dary <franck.dary@lis-lab.fr>
Date: Wed, 3 Mar 2021 15:03:27 +0100
Subject: [PATCH] Removed state from neuralnetwork
---
decoder/src/Beam.cpp | 7 +------
decoder/src/Decoder.cpp | 1 -
reading_machine/include/Classifier.hpp | 6 ++----
reading_machine/src/Classifier.cpp | 10 ++--------
reading_machine/src/ReadingMachine.cpp | 2 +-
torch_modules/include/ModularNetwork.hpp | 3 +--
torch_modules/include/NeuralNetwork.hpp | 9 ++-------
torch_modules/include/RandomNetwork.hpp | 2 +-
torch_modules/include/Submodule.hpp | 3 +--
torch_modules/src/ModularNetwork.cpp | 11 ++---------
torch_modules/src/RandomNetwork.cpp | 4 ++--
trainer/src/Trainer.cpp | 12 +++---------
12 files changed, 18 insertions(+), 52 deletions(-)
diff --git a/decoder/src/Beam.cpp b/decoder/src/Beam.cpp
index e39593c..9aad4f0 100644
--- a/decoder/src/Beam.cpp
+++ b/decoder/src/Beam.cpp
@@ -39,8 +39,6 @@ void Beam::update(ReadingMachine & machine, bool debug)
auto & classifier = *machine.getClassifier(elements[index].config.getState());
- classifier.setState(elements[index].config.getState());
-
if (machine.hasSplitWordTransitionSet())
elements[index].config.setAppliableSplitTransitions(machine.getSplitWordTransitionSet().getNAppliableTransitions(elements[index].config, Config::maxNbAppliableSplitTransitions));
@@ -50,7 +48,7 @@ void Beam::update(ReadingMachine & machine, bool debug)
auto context = classifier.getNN()->extractContext(elements[index].config).back();
auto neuralInput = torch::from_blob(context.data(), {(long)context.size()}, torch::TensorOptions(torch::kLong)).clone().to(NeuralNetworkImpl::device);
- auto prediction = classifier.isRegression() ? classifier.getNN()(neuralInput).squeeze(0) : torch::softmax(classifier.getNN()(neuralInput).squeeze(0), 0);
+ auto prediction = classifier.isRegression() ? classifier.getNN()->forward(neuralInput, elements[index].config.getState()).squeeze(0) : torch::softmax(classifier.getNN()->forward(neuralInput, elements[index].config.getState()).squeeze(0), 0);
float entropy = classifier.isRegression() ? 0.0 : NeuralNetworkImpl::entropy(prediction);
std::vector<std::pair<float, int>> scoresOfTransitions;
for (unsigned int i = 0; i < prediction.size(0); i++)
@@ -123,9 +121,6 @@ void Beam::update(ReadingMachine & machine, bool debug)
continue;
auto & config = element.config;
- auto & classifier = *machine.getClassifier(config.getState());
-
- classifier.setState(config.getState());
auto * transition = machine.getTransitionSet(config.getState()).getTransition(element.nextTransition);
diff --git a/decoder/src/Decoder.cpp b/decoder/src/Decoder.cpp
index 474db07..ad2e6a6 100644
--- a/decoder/src/Decoder.cpp
+++ b/decoder/src/Decoder.cpp
@@ -39,7 +39,6 @@ void Decoder::decode(BaseConfig & baseConfig, std::size_t beamSize, float beamTh
} catch(std::exception & e) {util::myThrow(e.what());}
baseConfig = beam[0].config;
- machine.getClassifier(baseConfig.getState())->setState(baseConfig.getState());
if (machine.getTransitionSet(baseConfig.getState()).getTransition("EOS b.0") and baseConfig.getLastNotEmptyHypConst(Config::EOSColName, baseConfig.getWordIndex()) != Config::EOSSymbol1)
{
diff --git a/reading_machine/include/Classifier.hpp b/reading_machine/include/Classifier.hpp
index 17bfc84..e4b2208 100644
--- a/reading_machine/include/Classifier.hpp
+++ b/reading_machine/include/Classifier.hpp
@@ -22,7 +22,6 @@ class Classifier
std::shared_ptr<NeuralNetworkImpl> nn;
std::unique_ptr<torch::optim::Optimizer> optimizer;
std::string optimizerType, optimizerParameters;
- std::string state;
std::vector<std::string> states;
std::filesystem::path path;
bool regression{false};
@@ -39,7 +38,7 @@ class Classifier
public :
Classifier(const std::string & name, std::filesystem::path path, std::vector<std::string> definition, bool train);
- TransitionSet & getTransitionSet();
+ TransitionSet & getTransitionSet(const std::string & state);
NeuralNetwork & getNN();
const std::string & getName() const;
int getNbParameters() const;
@@ -47,8 +46,7 @@ class Classifier
void loadOptimizer();
void saveOptimizer();
torch::optim::Optimizer & getOptimizer();
- void setState(const std::string & state);
- float getLossMultiplier();
+ float getLossMultiplier(const std::string & state);
const std::vector<std::string> & getStates() const;
void saveDicts();
void saveBest();
diff --git a/reading_machine/src/Classifier.cpp b/reading_machine/src/Classifier.cpp
index b5e929b..68e89b7 100644
--- a/reading_machine/src/Classifier.cpp
+++ b/reading_machine/src/Classifier.cpp
@@ -110,7 +110,7 @@ int Classifier::getNbParameters() const
return nbParameters;
}
-TransitionSet & Classifier::getTransitionSet()
+TransitionSet & Classifier::getTransitionSet(const std::string & state)
{
if (!transitionSets.count(state))
util::myThrow(fmt::format("cannot find transition set for state '{}'", state));
@@ -196,12 +196,6 @@ torch::optim::Optimizer & Classifier::getOptimizer()
return *optimizer;
}
-void Classifier::setState(const std::string & state)
-{
- this->state = state;
- nn->setState(state);
-}
-
void Classifier::initModular(const std::vector<std::string> & definition, std::size_t & curIndex, const std::map<std::string,std::size_t> & nbOutputsPerState)
{
std::string anyBlanks = "(?:(?:\\s|\\t)*)";
@@ -244,7 +238,7 @@ void Classifier::resetOptimizer()
util::myThrow(expected);
}
-float Classifier::getLossMultiplier()
+float Classifier::getLossMultiplier(const std::string & state)
{
return lossMultipliers.at(state);
}
diff --git a/reading_machine/src/ReadingMachine.cpp b/reading_machine/src/ReadingMachine.cpp
index 4336d3a..582acff 100644
--- a/reading_machine/src/ReadingMachine.cpp
+++ b/reading_machine/src/ReadingMachine.cpp
@@ -95,7 +95,7 @@ void ReadingMachine::readFromFile(std::filesystem::path path)
TransitionSet & ReadingMachine::getTransitionSet(const std::string & state)
{
- return classifiers[state2classifier.at(state)]->getTransitionSet();
+ return classifiers[state2classifier.at(state)]->getTransitionSet(state);
}
bool ReadingMachine::hasSplitWordTransitionSet() const
diff --git a/torch_modules/include/ModularNetwork.hpp b/torch_modules/include/ModularNetwork.hpp
index 9b7efae..ed73c30 100644
--- a/torch_modules/include/ModularNetwork.hpp
+++ b/torch_modules/include/ModularNetwork.hpp
@@ -29,7 +29,7 @@ class ModularNetworkImpl : public NeuralNetworkImpl
public :
ModularNetworkImpl(std::string name, std::map<std::string,std::size_t> nbOutputsPerState, std::vector<std::string> definitions, std::filesystem::path path);
- torch::Tensor forward(torch::Tensor input) override;
+ torch::Tensor forward(torch::Tensor input, const std::string & state) override;
std::vector<std::vector<long>> extractContext(Config & config) override;
void registerEmbeddings() override;
void saveDicts(std::filesystem::path path) override;
@@ -37,7 +37,6 @@ class ModularNetworkImpl : public NeuralNetworkImpl
void setDictsState(Dict::State state) override;
void setCountOcc(bool countOcc) override;
void removeRareDictElements(float rarityThreshold) override;
- void setState(const std::string & state);
};
#endif
diff --git a/torch_modules/include/NeuralNetwork.hpp b/torch_modules/include/NeuralNetwork.hpp
index 6058ceb..8215ad2 100644
--- a/torch_modules/include/NeuralNetwork.hpp
+++ b/torch_modules/include/NeuralNetwork.hpp
@@ -5,21 +5,16 @@
#include <filesystem>
#include "Config.hpp"
#include "NameHolder.hpp"
-#include "StateHolder.hpp"
-class NeuralNetworkImpl : public torch::nn::Module, public NameHolder, public StateHolder
+class NeuralNetworkImpl : public torch::nn::Module, public NameHolder
{
public :
static torch::Device device;
- private :
-
- std::string state;
-
public :
- virtual torch::Tensor forward(torch::Tensor input) = 0;
+ virtual torch::Tensor forward(torch::Tensor input, const std::string & state) = 0;
virtual std::vector<std::vector<long>> extractContext(Config & config) = 0;
virtual void registerEmbeddings() = 0;
virtual void saveDicts(std::filesystem::path path) = 0;
diff --git a/torch_modules/include/RandomNetwork.hpp b/torch_modules/include/RandomNetwork.hpp
index b20a779..3c559e9 100644
--- a/torch_modules/include/RandomNetwork.hpp
+++ b/torch_modules/include/RandomNetwork.hpp
@@ -12,7 +12,7 @@ class RandomNetworkImpl : public NeuralNetworkImpl
public :
RandomNetworkImpl(std::string name, std::map<std::string,std::size_t> nbOutputsPerState);
- torch::Tensor forward(torch::Tensor input) override;
+ torch::Tensor forward(torch::Tensor input, const std::string & state) override;
std::vector<std::vector<long>> extractContext(Config &) override;
void registerEmbeddings() override;
void saveDicts(std::filesystem::path path) override;
diff --git a/torch_modules/include/Submodule.hpp b/torch_modules/include/Submodule.hpp
index 553da4f..1dbbdc7 100644
--- a/torch_modules/include/Submodule.hpp
+++ b/torch_modules/include/Submodule.hpp
@@ -5,9 +5,8 @@
#include <filesystem>
#include "Config.hpp"
#include "DictHolder.hpp"
-#include "StateHolder.hpp"
-class Submodule : public torch::nn::Module, public DictHolder, public StateHolder
+class Submodule : public torch::nn::Module, public DictHolder
{
private :
diff --git a/torch_modules/src/ModularNetwork.cpp b/torch_modules/src/ModularNetwork.cpp
index e2e225c..c936f85 100644
--- a/torch_modules/src/ModularNetwork.cpp
+++ b/torch_modules/src/ModularNetwork.cpp
@@ -80,7 +80,7 @@ ModularNetworkImpl::ModularNetworkImpl(std::string name, std::map<std::string,st
outputLayersPerState.emplace(it.first,register_module(fmt::format("output_{}",it.first), torch::nn::Linear(mlp->outputSize(), it.second)));
}
-torch::Tensor ModularNetworkImpl::forward(torch::Tensor input)
+torch::Tensor ModularNetworkImpl::forward(torch::Tensor input, const std::string & state)
{
if (input.dim() == 1)
input = input.unsqueeze(0);
@@ -92,7 +92,7 @@ torch::Tensor ModularNetworkImpl::forward(torch::Tensor input)
auto totalInput = inputDropout(torch::cat(outputs, 1));
- return outputLayersPerState.at(getState())(mlp(totalInput));
+ return outputLayersPerState.at(state)(mlp(totalInput));
}
std::vector<std::vector<long>> ModularNetworkImpl::extractContext(Config & config)
@@ -149,10 +149,3 @@ void ModularNetworkImpl::removeRareDictElements(float rarityThreshold)
}
}
-void ModularNetworkImpl::setState(const std::string & state)
-{
- NeuralNetworkImpl::setState(state);
- for (auto & mod : modules)
- mod->setState(state);
-}
-
diff --git a/torch_modules/src/RandomNetwork.cpp b/torch_modules/src/RandomNetwork.cpp
index 7a6491b..87a6046 100644
--- a/torch_modules/src/RandomNetwork.cpp
+++ b/torch_modules/src/RandomNetwork.cpp
@@ -5,12 +5,12 @@ RandomNetworkImpl::RandomNetworkImpl(std::string name, std::map<std::string,std:
setName(name);
}
-torch::Tensor RandomNetworkImpl::forward(torch::Tensor input)
+torch::Tensor RandomNetworkImpl::forward(torch::Tensor input, const std::string & state)
{
if (input.dim() == 1)
input = input.unsqueeze(0);
- return torch::randn({input.size(0), (long)nbOutputsPerState[getState()]}, torch::TensorOptions().device(device).requires_grad(true));
+ return torch::randn({input.size(0), (long)nbOutputsPerState[state]}, torch::TensorOptions().device(device).requires_grad(true));
}
std::vector<std::vector<long>> RandomNetworkImpl::extractContext(Config &)
diff --git a/trainer/src/Trainer.cpp b/trainer/src/Trainer.cpp
index 7f1aaec..56ecc44 100644
--- a/trainer/src/Trainer.cpp
+++ b/trainer/src/Trainer.cpp
@@ -53,7 +53,6 @@ void Trainer::extractExamples(std::vector<SubConfig> & configs, bool debug, std:
config.addPredicted(machine.getPredicted());
config.setStrategy(machine.getStrategyDefinition());
config.setState(config.getStrategy().getInitialState());
- machine.getClassifier(config.getState())->setState(config.getState());
while (true)
{
@@ -94,7 +93,7 @@ void Trainer::extractExamples(std::vector<SubConfig> & configs, bool debug, std:
{
auto & classifier = *machine.getClassifier(config.getState());
auto neuralInput = torch::from_blob(context.data(), {(long)context.size()}, torch::TensorOptions(torch::kLong)).clone().to(NeuralNetworkImpl::device);
- auto prediction = classifier.isRegression() ? classifier.getNN()(neuralInput).squeeze(0) : torch::softmax(classifier.getNN()(neuralInput).squeeze(0), 0);
+ auto prediction = classifier.isRegression() ? classifier.getNN()->forward(neuralInput, config.getState()).squeeze(0) : torch::softmax(classifier.getNN()->forward(neuralInput, config.getState()).squeeze(0), 0);
entropy = NeuralNetworkImpl::entropy(prediction);
std::vector<int> candidates;
@@ -176,7 +175,6 @@ void Trainer::extractExamples(std::vector<SubConfig> & configs, bool debug, std:
break;
config.setState(movement.first);
- machine.getClassifier(config.getState())->setState(movement.first);
config.moveWordIndexRelaxed(movement.second);
if (config.needsUpdate())
@@ -217,9 +215,7 @@ float Trainer::processDataset(DataLoader & loader, bool train, bool printAdvance
if (train)
machine.getClassifier(state)->getOptimizer().zero_grad();
- machine.getClassifier(state)->setState(state);
-
- auto prediction = machine.getClassifier(state)->getNN()(data);
+ auto prediction = machine.getClassifier(state)->getNN()->forward(data, state);
if (prediction.dim() == 1)
prediction = prediction.unsqueeze(0);
@@ -229,7 +225,7 @@ float Trainer::processDataset(DataLoader & loader, bool train, bool printAdvance
labels /= util::float2longScale;
}
- auto loss = machine.getClassifier(state)->getLossMultiplier()*machine.getClassifier(state)->getLossFunction()(prediction, labels);
+ auto loss = machine.getClassifier(state)->getLossMultiplier(state)*machine.getClassifier(state)->getLossFunction()(prediction, labels);
float lossAsFloat = 0.0;
try
{
@@ -340,7 +336,6 @@ void Trainer::extractActionSequence(BaseConfig & config)
config.addPredicted(machine.getPredicted());
config.setStrategy(machine.getStrategyDefinition());
config.setState(config.getStrategy().getInitialState());
- machine.getClassifier(config.getState())->setState(config.getState());
int curSeq = 0;
int curSeqStartIndex = -1;
@@ -403,7 +398,6 @@ void Trainer::extractActionSequence(BaseConfig & config)
break;
config.setState(movement.first);
- machine.getClassifier(config.getState())->setState(movement.first);
config.moveWordIndexRelaxed(movement.second);
}
--
GitLab