From 11e87ce1fd20636526e6259ffbad18883b42fce5 Mon Sep 17 00:00:00 2001
From: Franck Dary <franck.dary@lis-lab.fr>
Date: Wed, 3 Mar 2021 16:26:42 +0100
Subject: [PATCH] extractContext now directly gives a torch::Tensor
---
decoder/src/Beam.cpp | 3 +-
.../include/AppliableTransModule.hpp | 2 +-
torch_modules/include/ContextModule.hpp | 2 +-
torch_modules/include/ContextualModule.hpp | 2 +-
.../include/DepthLayerTreeEmbeddingModule.hpp | 2 +-
torch_modules/include/DistanceModule.hpp | 2 +-
torch_modules/include/FocusedColumnModule.hpp | 2 +-
torch_modules/include/HistoryModule.hpp | 2 +-
torch_modules/include/ModularNetwork.hpp | 3 +-
torch_modules/include/NeuralNetwork.hpp | 2 +-
torch_modules/include/NumericColumnModule.hpp | 2 +-
torch_modules/include/RandomNetwork.hpp | 2 +-
torch_modules/include/RawInputModule.hpp | 2 +-
torch_modules/include/SplitTransModule.hpp | 2 +-
torch_modules/include/StateNameModule.hpp | 2 +-
torch_modules/include/Submodule.hpp | 2 +-
torch_modules/include/UppercaseRateModule.hpp | 2 +-
torch_modules/src/AppliableTransModule.cpp | 11 +-
torch_modules/src/ContextModule.cpp | 16 ++-
torch_modules/src/ContextualModule.cpp | 44 ++++----
.../src/DepthLayerTreeEmbeddingModule.cpp | 48 ++++----
torch_modules/src/DistanceModule.cpp | 29 ++---
torch_modules/src/FocusedColumnModule.cpp | 104 +++++++++---------
torch_modules/src/HistoryModule.cpp | 13 +--
torch_modules/src/ModularNetwork.cpp | 6 +-
torch_modules/src/NumericColumnModule.cpp | 28 ++---
torch_modules/src/RandomNetwork.cpp | 5 +-
torch_modules/src/RawInputModule.cpp | 40 ++++---
torch_modules/src/SplitTransModule.cpp | 13 +--
torch_modules/src/StateNameModule.cpp | 5 +-
torch_modules/src/UppercaseRateModule.cpp | 33 +++---
trainer/include/Trainer.hpp | 2 +-
trainer/src/Trainer.cpp | 14 +--
33 files changed, 226 insertions(+), 221 deletions(-)
diff --git a/decoder/src/Beam.cpp b/decoder/src/Beam.cpp
index 9aad4f0..c932733 100644
--- a/decoder/src/Beam.cpp
+++ b/decoder/src/Beam.cpp
@@ -45,8 +45,7 @@ void Beam::update(ReadingMachine & machine, bool debug)
auto appliableTransitions = machine.getTransitionSet(elements[index].config.getState()).getAppliableTransitions(elements[index].config);
elements[index].config.setAppliableTransitions(appliableTransitions);
- 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 neuralInput = classifier.getNN()->extractContext(elements[index].config);
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);
diff --git a/torch_modules/include/AppliableTransModule.hpp b/torch_modules/include/AppliableTransModule.hpp
index 5e6f9e4..98f5fe1 100644
--- a/torch_modules/include/AppliableTransModule.hpp
+++ b/torch_modules/include/AppliableTransModule.hpp
@@ -19,7 +19,7 @@ class AppliableTransModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(AppliableTransModule);
diff --git a/torch_modules/include/ContextModule.hpp b/torch_modules/include/ContextModule.hpp
index 5851887..c2e0668 100644
--- a/torch_modules/include/ContextModule.hpp
+++ b/torch_modules/include/ContextModule.hpp
@@ -30,7 +30,7 @@ class ContextModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(ContextModule);
diff --git a/torch_modules/include/ContextualModule.hpp b/torch_modules/include/ContextualModule.hpp
index e7fb2a9..8483b1a 100644
--- a/torch_modules/include/ContextualModule.hpp
+++ b/torch_modules/include/ContextualModule.hpp
@@ -31,7 +31,7 @@ class ContextualModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(ContextualModule);
diff --git a/torch_modules/include/DepthLayerTreeEmbeddingModule.hpp b/torch_modules/include/DepthLayerTreeEmbeddingModule.hpp
index 6da8943..3621e6e 100644
--- a/torch_modules/include/DepthLayerTreeEmbeddingModule.hpp
+++ b/torch_modules/include/DepthLayerTreeEmbeddingModule.hpp
@@ -27,7 +27,7 @@ class DepthLayerTreeEmbeddingModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(DepthLayerTreeEmbeddingModule);
diff --git a/torch_modules/include/DistanceModule.hpp b/torch_modules/include/DistanceModule.hpp
index 3702ad5..bafa0b8 100644
--- a/torch_modules/include/DistanceModule.hpp
+++ b/torch_modules/include/DistanceModule.hpp
@@ -26,7 +26,7 @@ class DistanceModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(DistanceModule);
diff --git a/torch_modules/include/FocusedColumnModule.hpp b/torch_modules/include/FocusedColumnModule.hpp
index af370c6..a7df331 100644
--- a/torch_modules/include/FocusedColumnModule.hpp
+++ b/torch_modules/include/FocusedColumnModule.hpp
@@ -29,7 +29,7 @@ class FocusedColumnModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(FocusedColumnModule);
diff --git a/torch_modules/include/HistoryModule.hpp b/torch_modules/include/HistoryModule.hpp
index 54418a6..b4a725b 100644
--- a/torch_modules/include/HistoryModule.hpp
+++ b/torch_modules/include/HistoryModule.hpp
@@ -25,7 +25,7 @@ class HistoryModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(HistoryModule);
diff --git a/torch_modules/include/ModularNetwork.hpp b/torch_modules/include/ModularNetwork.hpp
index ed73c30..31685e2 100644
--- a/torch_modules/include/ModularNetwork.hpp
+++ b/torch_modules/include/ModularNetwork.hpp
@@ -25,12 +25,13 @@ class ModularNetworkImpl : public NeuralNetworkImpl
MLP mlp{nullptr};
std::vector<std::shared_ptr<Submodule>> modules;
std::map<std::string,torch::nn::Linear> outputLayersPerState;
+ std::size_t totalInputSize{0};
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, const std::string & state) override;
- std::vector<std::vector<long>> extractContext(Config & config) override;
+ torch::Tensor extractContext(Config & config) override;
void registerEmbeddings() override;
void saveDicts(std::filesystem::path path) override;
void loadDicts(std::filesystem::path path) override;
diff --git a/torch_modules/include/NeuralNetwork.hpp b/torch_modules/include/NeuralNetwork.hpp
index 8215ad2..ffbcdea 100644
--- a/torch_modules/include/NeuralNetwork.hpp
+++ b/torch_modules/include/NeuralNetwork.hpp
@@ -15,7 +15,7 @@ class NeuralNetworkImpl : public torch::nn::Module, public NameHolder
public :
virtual torch::Tensor forward(torch::Tensor input, const std::string & state) = 0;
- virtual std::vector<std::vector<long>> extractContext(Config & config) = 0;
+ virtual torch::Tensor extractContext(Config & config) = 0;
virtual void registerEmbeddings() = 0;
virtual void saveDicts(std::filesystem::path path) = 0;
virtual void loadDicts(std::filesystem::path path) = 0;
diff --git a/torch_modules/include/NumericColumnModule.hpp b/torch_modules/include/NumericColumnModule.hpp
index e3d46ae..3ee9cb2 100644
--- a/torch_modules/include/NumericColumnModule.hpp
+++ b/torch_modules/include/NumericColumnModule.hpp
@@ -24,7 +24,7 @@ class NumericColumnModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(NumericColumnModule);
diff --git a/torch_modules/include/RandomNetwork.hpp b/torch_modules/include/RandomNetwork.hpp
index 3c559e9..33d99a1 100644
--- a/torch_modules/include/RandomNetwork.hpp
+++ b/torch_modules/include/RandomNetwork.hpp
@@ -13,7 +13,7 @@ class RandomNetworkImpl : public NeuralNetworkImpl
RandomNetworkImpl(std::string name, std::map<std::string,std::size_t> nbOutputsPerState);
torch::Tensor forward(torch::Tensor input, const std::string & state) override;
- std::vector<std::vector<long>> extractContext(Config &) override;
+ torch::Tensor extractContext(Config &) override;
void registerEmbeddings() override;
void saveDicts(std::filesystem::path path) override;
void loadDicts(std::filesystem::path path) override;
diff --git a/torch_modules/include/RawInputModule.hpp b/torch_modules/include/RawInputModule.hpp
index 26237e2..0ca658b 100644
--- a/torch_modules/include/RawInputModule.hpp
+++ b/torch_modules/include/RawInputModule.hpp
@@ -25,7 +25,7 @@ class RawInputModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(RawInputModule);
diff --git a/torch_modules/include/SplitTransModule.hpp b/torch_modules/include/SplitTransModule.hpp
index 1ef1796..b88491e 100644
--- a/torch_modules/include/SplitTransModule.hpp
+++ b/torch_modules/include/SplitTransModule.hpp
@@ -24,7 +24,7 @@ class SplitTransModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(SplitTransModule);
diff --git a/torch_modules/include/StateNameModule.hpp b/torch_modules/include/StateNameModule.hpp
index 3abfe82..ace1cbc 100644
--- a/torch_modules/include/StateNameModule.hpp
+++ b/torch_modules/include/StateNameModule.hpp
@@ -21,7 +21,7 @@ class StateNameModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(StateNameModule);
diff --git a/torch_modules/include/Submodule.hpp b/torch_modules/include/Submodule.hpp
index 1dbbdc7..f4722bf 100644
--- a/torch_modules/include/Submodule.hpp
+++ b/torch_modules/include/Submodule.hpp
@@ -24,7 +24,7 @@ class Submodule : public torch::nn::Module, public DictHolder
void loadPretrainedW2vEmbeddings(torch::nn::Embedding embeddings, std::filesystem::path path, std::string prefix);
virtual std::size_t getOutputSize() = 0;
virtual std::size_t getInputSize() = 0;
- virtual void addToContext(std::vector<std::vector<long>> & context, const Config & config) = 0;
+ virtual void addToContext(torch::Tensor & context, const Config & config) = 0;
virtual torch::Tensor forward(torch::Tensor input) = 0;
virtual void registerEmbeddings() = 0;
std::function<std::string(const std::string &)> getFunction(const std::string functionNames);
diff --git a/torch_modules/include/UppercaseRateModule.hpp b/torch_modules/include/UppercaseRateModule.hpp
index dcfb89c..9495661 100644
--- a/torch_modules/include/UppercaseRateModule.hpp
+++ b/torch_modules/include/UppercaseRateModule.hpp
@@ -22,7 +22,7 @@ class UppercaseRateModuleImpl : public Submodule
torch::Tensor forward(torch::Tensor input);
std::size_t getOutputSize() override;
std::size_t getInputSize() override;
- void addToContext(std::vector<std::vector<long>> & context, const Config & config) override;
+ void addToContext(torch::Tensor & context, const Config & config) override;
void registerEmbeddings() override;
};
TORCH_MODULE(UppercaseRateModule);
diff --git a/torch_modules/src/AppliableTransModule.cpp b/torch_modules/src/AppliableTransModule.cpp
index c50586f..7a5c830 100644
--- a/torch_modules/src/AppliableTransModule.cpp
+++ b/torch_modules/src/AppliableTransModule.cpp
@@ -20,15 +20,12 @@ std::size_t AppliableTransModuleImpl::getInputSize()
return nbTrans;
}
-void AppliableTransModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void AppliableTransModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
auto & appliableTrans = config.getAppliableTransitions();
- for (auto & contextElement : context)
- for (int i = 0; i < nbTrans; i++)
- if (i < (int)appliableTrans.size())
- contextElement.emplace_back(appliableTrans[i]);
- else
- contextElement.emplace_back(0);
+ for (int i = 0; i < nbTrans; i++)
+ if (i < (int)appliableTrans.size())
+ context[firstInputIndex+i] = appliableTrans[i];
}
void AppliableTransModuleImpl::registerEmbeddings()
diff --git a/torch_modules/src/ContextModule.cpp b/torch_modules/src/ContextModule.cpp
index f457f31..19c7972 100644
--- a/torch_modules/src/ContextModule.cpp
+++ b/torch_modules/src/ContextModule.cpp
@@ -83,7 +83,7 @@ std::size_t ContextModuleImpl::getInputSize()
return columns.size()*(targets.size());
}
-void ContextModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void ContextModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
auto & dict = getDict();
std::vector<long> contextIndexes;
@@ -125,24 +125,22 @@ void ContextModuleImpl::addToContext(std::vector<std::vector<long>> & context, c
contextIndexes.emplace_back(-3);
}
+ int insertIndex = 0;
for (auto index : contextIndexes)
for (unsigned int colIndex = 0; colIndex < columns.size(); colIndex++)
{
auto & col = columns[colIndex];
if (index == -1)
{
- for (auto & contextElement : context)
- contextElement.push_back(dict.getIndexOrInsert(Dict::nullValueStr, col));
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::nullValueStr, col);
}
else if (index == -2)
{
- for (auto & contextElement : context)
- contextElement.push_back(dict.getIndexOrInsert(Dict::noChildValueStr, col));
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::noChildValueStr, col);
}
else if (index == -3)
{
- for (auto & contextElement : context)
- contextElement.push_back(dict.getIndexOrInsert(Dict::oobValueStr, col));
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::oobValueStr, col);
}
else
{
@@ -162,9 +160,9 @@ void ContextModuleImpl::addToContext(std::vector<std::vector<long>> & context, c
dictIndex = dict.getIndexOrInsert(featureValue, col);
}
- for (auto & contextElement : context)
- contextElement.push_back(dictIndex);
+ context[firstInputIndex+insertIndex] = dictIndex;
}
+ insertIndex++;
}
}
diff --git a/torch_modules/src/ContextualModule.cpp b/torch_modules/src/ContextualModule.cpp
index f5f8562..1c56983 100644
--- a/torch_modules/src/ContextualModule.cpp
+++ b/torch_modules/src/ContextualModule.cpp
@@ -87,7 +87,7 @@ std::size_t ContextualModuleImpl::getInputSize()
return columns.size()*(4+window.second-window.first)+targets.size();
}
-void ContextualModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void ContextualModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
auto & dict = getDict();
std::vector<long> contextIndexes;
@@ -132,24 +132,23 @@ void ContextualModuleImpl::addToContext(std::vector<std::vector<long>> & context
else
targetIndexes.emplace_back(-1);
+ int insertIndex = 0;
+
for (auto index : contextIndexes)
for (unsigned int colIndex = 0; colIndex < columns.size(); colIndex++)
{
auto & col = columns[colIndex];
if (index == -1)
{
- for (auto & contextElement : context)
- contextElement.push_back(dict.getIndexOrInsert(Dict::nullValueStr, col));
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::nullValueStr, col);
}
else if (index == -2)
{
- for (auto & contextElement : context)
- contextElement.push_back(dict.getIndexOrInsert(Dict::noChildValueStr, col));
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::noChildValueStr, col);
}
else if (index == -3)
{
- for (auto & contextElement : context)
- contextElement.push_back(dict.getIndexOrInsert(Dict::oobValueStr, col));
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::oobValueStr, col);
}
else
{
@@ -169,33 +168,32 @@ void ContextualModuleImpl::addToContext(std::vector<std::vector<long>> & context
dictIndex = dict.getIndexOrInsert(functions[colIndex](featureValue), col);
}
- for (auto & contextElement : context)
- contextElement.push_back(dictIndex);
+ context[firstInputIndex+insertIndex] = dictIndex;
}
+
+ insertIndex++;
}
for (auto index : targetIndexes)
{
if (configIndex2ContextIndex.count(index))
{
- for (auto & contextElement : context)
- contextElement.push_back(configIndex2ContextIndex.at(index));
+ context[firstInputIndex+insertIndex] = configIndex2ContextIndex.at(index);
}
else
{
- for (auto & contextElement : context)
- {
- // -1 == doesn't exist (s.0 when no stack)
- if (index == -1)
- contextElement.push_back(0);
- // -2 == nochild
- else if (index == -2)
- contextElement.push_back(1);
- // other == out of context bounds
- else
- contextElement.push_back(2);
- }
+ // -1 == doesn't exist (s.0 when no stack)
+ if (index == -1)
+ context[firstInputIndex+insertIndex] = 0;
+ // -2 == nochild
+ else if (index == -2)
+ context[firstInputIndex+insertIndex] = 1;
+ // other == out of context bounds
+ else
+ context[firstInputIndex+insertIndex] = 2;
}
+
+ insertIndex++;
}
}
diff --git a/torch_modules/src/DepthLayerTreeEmbeddingModule.cpp b/torch_modules/src/DepthLayerTreeEmbeddingModule.cpp
index 06c0b5f..ac90690 100644
--- a/torch_modules/src/DepthLayerTreeEmbeddingModule.cpp
+++ b/torch_modules/src/DepthLayerTreeEmbeddingModule.cpp
@@ -84,7 +84,7 @@ std::size_t DepthLayerTreeEmbeddingModuleImpl::getInputSize()
return inputSize;
}
-void DepthLayerTreeEmbeddingModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void DepthLayerTreeEmbeddingModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
auto & dict = getDict();
std::vector<long> focusedIndexes;
@@ -98,30 +98,34 @@ void DepthLayerTreeEmbeddingModuleImpl::addToContext(std::vector<std::vector<lon
else
focusedIndexes.emplace_back(-1);
- for (auto & contextElement : context)
- for (auto index : focusedIndexes)
+ int insertIndex = 0;
+ for (auto index : focusedIndexes)
+ {
+ std::vector<std::string> childs{std::to_string(index)};
+
+ for (unsigned int depth = 0; depth < maxElemPerDepth.size(); depth++)
{
- std::vector<std::string> childs{std::to_string(index)};
+ std::vector<std::string> newChilds;
+ for (auto & child : childs)
+ if (config.has(Config::childsColName, std::stoi(child), 0))
+ {
+ auto val = util::split(config.getAsFeature(Config::childsColName, std::stoi(child)).get(), '|');
+ newChilds.insert(newChilds.end(), val.begin(), val.end());
+ }
+ childs = newChilds;
- for (unsigned int depth = 0; depth < maxElemPerDepth.size(); depth++)
- {
- std::vector<std::string> newChilds;
- for (auto & child : childs)
- if (config.has(Config::childsColName, std::stoi(child), 0))
- {
- auto val = util::split(config.getAsFeature(Config::childsColName, std::stoi(child)).get(), '|');
- newChilds.insert(newChilds.end(), val.begin(), val.end());
- }
- childs = newChilds;
-
- for (int i = 0; i < maxElemPerDepth[depth]; i++)
- for (auto & col : columns)
- if (i < (int)newChilds.size() and config.has(col, std::stoi(newChilds[i]), 0))
- contextElement.emplace_back(dict.getIndexOrInsert(config.getAsFeature(col,std::stoi(newChilds[i])), col));
- else
- contextElement.emplace_back(dict.getIndexOrInsert(Dict::nullValueStr, col));
- }
+ for (int i = 0; i < maxElemPerDepth[depth]; i++)
+ for (auto & col : columns)
+ {
+ if (i < (int)newChilds.size() and config.has(col, std::stoi(newChilds[i]), 0))
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(config.getAsFeature(col,std::stoi(newChilds[i])), col);
+ else
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::nullValueStr, col);
+
+ insertIndex++;
+ }
}
+ }
}
void DepthLayerTreeEmbeddingModuleImpl::registerEmbeddings()
diff --git a/torch_modules/src/DistanceModule.cpp b/torch_modules/src/DistanceModule.cpp
index 2e71e25..fddf2e0 100644
--- a/torch_modules/src/DistanceModule.cpp
+++ b/torch_modules/src/DistanceModule.cpp
@@ -63,7 +63,7 @@ std::size_t DistanceModuleImpl::getInputSize()
return (fromBuffer.size()+fromStack.size()) * (toBuffer.size()+toStack.size());
}
-void DistanceModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void DistanceModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
auto & dict = getDict();
std::vector<long> fromIndexes, toIndexes;
@@ -88,25 +88,26 @@ void DistanceModuleImpl::addToContext(std::vector<std::vector<long>> & context,
std::string prefix = "DISTANCE";
- for (auto & contextElement : context)
- {
- for (auto from : fromIndexes)
- for (auto to : toIndexes)
+ int insertIndex = 0;
+ for (auto from : fromIndexes)
+ for (auto to : toIndexes)
+ {
+ if (from == -1 or to == -1)
+ {
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::nullValueStr, prefix);
+ }
+ else
{
- if (from == -1 or to == -1)
- {
- contextElement.emplace_back(dict.getIndexOrInsert(Dict::nullValueStr, prefix));
- continue;
- }
-
long dist = std::abs(config.getRelativeDistance(from, to));
if (dist <= threshold)
- contextElement.emplace_back(dict.getIndexOrInsert(fmt::format("{}({})", prefix, dist), ""));
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(fmt::format("{}({})", prefix, dist), "");
else
- contextElement.emplace_back(dict.getIndexOrInsert(Dict::unknownValueStr, prefix));
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::unknownValueStr, prefix);
}
- }
+
+ insertIndex++;
+ }
}
void DistanceModuleImpl::registerEmbeddings()
diff --git a/torch_modules/src/FocusedColumnModule.cpp b/torch_modules/src/FocusedColumnModule.cpp
index 115f918..3fc25f0 100644
--- a/torch_modules/src/FocusedColumnModule.cpp
+++ b/torch_modules/src/FocusedColumnModule.cpp
@@ -82,7 +82,7 @@ std::size_t FocusedColumnModuleImpl::getInputSize()
return (focusedBuffer.size()+focusedStack.size()) * maxNbElements;
}
-void FocusedColumnModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void FocusedColumnModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
auto & dict = getDict();
std::vector<long> focusedIndexes;
@@ -96,63 +96,67 @@ void FocusedColumnModuleImpl::addToContext(std::vector<std::vector<long>> & cont
else
focusedIndexes.emplace_back(-1);
- for (auto & contextElement : context)
+ int insertIndex = 0;
+ for (auto index : focusedIndexes)
{
- for (auto index : focusedIndexes)
+ if (index == -1)
{
- if (index == -1)
+ for (int i = 0; i < maxNbElements; i++)
{
- for (int i = 0; i < maxNbElements; i++)
- contextElement.emplace_back(dict.getIndexOrInsert(Dict::nullValueStr, column));
- continue;
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::nullValueStr, column);
+ insertIndex++;
}
+ continue;
+ }
- std::vector<std::string> elements;
- if (column == "FORM")
- {
- auto asUtf8 = util::splitAsUtf8(func(config.getAsFeature(column, index).get()));
-
- //TODO don't use nullValueStr here
- for (int i = 0; i < maxNbElements; i++)
- if (i < (int)asUtf8.size())
- elements.emplace_back(fmt::format("{}", asUtf8[i]));
- else
- elements.emplace_back(Dict::nullValueStr);
- }
- else if (column == "FEATS")
- {
- auto splited = util::split(func(config.getAsFeature(column, index).get()), '|');
+ std::vector<std::string> elements;
+ if (column == "FORM")
+ {
+ auto asUtf8 = util::splitAsUtf8(func(config.getAsFeature(column, index).get()));
+
+ //TODO don't use nullValueStr here
+ for (int i = 0; i < maxNbElements; i++)
+ if (i < (int)asUtf8.size())
+ elements.emplace_back(fmt::format("{}", asUtf8[i]));
+ else
+ elements.emplace_back(Dict::nullValueStr);
+ }
+ else if (column == "FEATS")
+ {
+ auto splited = util::split(func(config.getAsFeature(column, index).get()), '|');
- for (int i = 0; i < maxNbElements; i++)
- if (i < (int)splited.size())
- elements.emplace_back(splited[i]);
- else
- elements.emplace_back(Dict::nullValueStr);
- }
- else if (column == "ID")
- {
- if (config.isTokenPredicted(index))
- elements.emplace_back("TOKEN");
- else if (config.isMultiwordPredicted(index))
- elements.emplace_back("MULTIWORD");
- else if (config.isEmptyNodePredicted(index))
- elements.emplace_back("EMPTYNODE");
- }
- else if (column == "EOS")
- {
- bool isEOS = func(config.getAsFeature(Config::EOSColName, index)) == Config::EOSSymbol1;
- elements.emplace_back(fmt::format("{}", isEOS));
- }
- else
- {
- elements.emplace_back(func(config.getAsFeature(column, index)));
- }
+ for (int i = 0; i < maxNbElements; i++)
+ if (i < (int)splited.size())
+ elements.emplace_back(splited[i]);
+ else
+ elements.emplace_back(Dict::nullValueStr);
+ }
+ else if (column == "ID")
+ {
+ if (config.isTokenPredicted(index))
+ elements.emplace_back("TOKEN");
+ else if (config.isMultiwordPredicted(index))
+ elements.emplace_back("MULTIWORD");
+ else if (config.isEmptyNodePredicted(index))
+ elements.emplace_back("EMPTYNODE");
+ }
+ else if (column == "EOS")
+ {
+ bool isEOS = func(config.getAsFeature(Config::EOSColName, index)) == Config::EOSSymbol1;
+ elements.emplace_back(fmt::format("{}", isEOS));
+ }
+ else
+ {
+ elements.emplace_back(func(config.getAsFeature(column, index)));
+ }
- if ((int)elements.size() != maxNbElements)
- util::myThrow(fmt::format("elements.size ({}) != maxNbElements ({})", elements.size(), maxNbElements));
+ if ((int)elements.size() != maxNbElements)
+ util::myThrow(fmt::format("elements.size ({}) != maxNbElements ({})", elements.size(), maxNbElements));
- for (auto & element : elements)
- contextElement.emplace_back(dict.getIndexOrInsert(element, column));
+ for (auto & element : elements)
+ {
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(element, column);
+ insertIndex++;
}
}
}
diff --git a/torch_modules/src/HistoryModule.cpp b/torch_modules/src/HistoryModule.cpp
index 7d0912c..4a9033f 100644
--- a/torch_modules/src/HistoryModule.cpp
+++ b/torch_modules/src/HistoryModule.cpp
@@ -53,18 +53,17 @@ std::size_t HistoryModuleImpl::getInputSize()
return maxNbElements;
}
-void HistoryModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void HistoryModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
auto & dict = getDict();
std::string prefix = "HISTORY";
- for (auto & contextElement : context)
- for (int i = 0; i < maxNbElements; i++)
- if (config.hasHistory(i))
- contextElement.emplace_back(dict.getIndexOrInsert(config.getHistory(i), prefix));
- else
- contextElement.emplace_back(dict.getIndexOrInsert(Dict::nullValueStr, prefix));
+ for (int i = 0; i < maxNbElements; i++)
+ if (config.hasHistory(i))
+ context[firstInputIndex+i] = dict.getIndexOrInsert(config.getHistory(i), prefix);
+ else
+ context[firstInputIndex+i] = dict.getIndexOrInsert(Dict::nullValueStr, prefix);
}
void HistoryModuleImpl::registerEmbeddings()
diff --git a/torch_modules/src/ModularNetwork.cpp b/torch_modules/src/ModularNetwork.cpp
index c936f85..1c39f18 100644
--- a/torch_modules/src/ModularNetwork.cpp
+++ b/torch_modules/src/ModularNetwork.cpp
@@ -69,6 +69,8 @@ ModularNetworkImpl::ModularNetworkImpl(std::string name, std::map<std::string,st
currentOutputSize += modules.back()->getOutputSize();
}
+ totalInputSize = currentInputSize;
+
if (mlpDef.empty())
util::myThrow("no MLP definition found");
if (inputDropout.is_empty())
@@ -95,9 +97,9 @@ torch::Tensor ModularNetworkImpl::forward(torch::Tensor input, const std::string
return outputLayersPerState.at(state)(mlp(totalInput));
}
-std::vector<std::vector<long>> ModularNetworkImpl::extractContext(Config & config)
+torch::Tensor ModularNetworkImpl::extractContext(Config & config)
{
- std::vector<std::vector<long>> context(1);
+ torch::Tensor context = torch::zeros({totalInputSize}, torch::TensorOptions(torch::kLong).device(NeuralNetworkImpl::device));
for (auto & mod : modules)
mod->addToContext(context, config);
return context;
diff --git a/torch_modules/src/NumericColumnModule.cpp b/torch_modules/src/NumericColumnModule.cpp
index 15d2b19..49d3016 100644
--- a/torch_modules/src/NumericColumnModule.cpp
+++ b/torch_modules/src/NumericColumnModule.cpp
@@ -60,7 +60,7 @@ std::size_t NumericColumnModuleImpl::getInputSize()
return focusedBuffer.size() + focusedStack.size();
}
-void NumericColumnModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void NumericColumnModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
std::vector<long> focusedIndexes;
@@ -73,21 +73,21 @@ void NumericColumnModuleImpl::addToContext(std::vector<std::vector<long>> & cont
else
focusedIndexes.emplace_back(-1);
- for (auto & contextElement : context)
- for (auto index : focusedIndexes)
+ int insertIndex = 0;
+ for (auto index : focusedIndexes)
+ {
+ double res = 0.0;
+ if (index >= 0)
{
- double res = 0.0;
- if (index >= 0)
- {
- auto value = config.getAsFeature(column, index).get();
- try {res = (value == "_" or value == "NA") ? defaultValue : std::stof(value);}
- catch (std::exception & e)
- {util::myThrow(fmt::format("{} for '{}'", e.what(), value));}
- }
-
- contextElement.emplace_back(0);
- std::memcpy(&contextElement.back(), &res, sizeof res);
+ auto value = config.getAsFeature(column, index).get();
+ try {res = (value == "_" or value == "NA") ? defaultValue : std::stof(value);}
+ catch (std::exception & e)
+ {util::myThrow(fmt::format("{} for '{}'", e.what(), value));}
}
+
+ //TODO : Check if this works
+ context[firstInputIndex+insertIndex] = res;
+ }
}
void NumericColumnModuleImpl::registerEmbeddings()
diff --git a/torch_modules/src/RandomNetwork.cpp b/torch_modules/src/RandomNetwork.cpp
index 87a6046..b05d1aa 100644
--- a/torch_modules/src/RandomNetwork.cpp
+++ b/torch_modules/src/RandomNetwork.cpp
@@ -13,9 +13,10 @@ torch::Tensor RandomNetworkImpl::forward(torch::Tensor input, const std::string
return torch::randn({input.size(0), (long)nbOutputsPerState[state]}, torch::TensorOptions().device(device).requires_grad(true));
}
-std::vector<std::vector<long>> RandomNetworkImpl::extractContext(Config &)
+torch::Tensor RandomNetworkImpl::extractContext(Config &)
{
- return std::vector<std::vector<long>>{{0}};
+ torch::Tensor context;
+ return context;
}
void RandomNetworkImpl::registerEmbeddings()
diff --git a/torch_modules/src/RawInputModule.cpp b/torch_modules/src/RawInputModule.cpp
index d948386..2d6bd62 100644
--- a/torch_modules/src/RawInputModule.cpp
+++ b/torch_modules/src/RawInputModule.cpp
@@ -11,6 +11,9 @@ RawInputModuleImpl::RawInputModuleImpl(std::string name, const std::string & def
leftWindow = std::stoi(sm.str(1));
rightWindow = std::stoi(sm.str(2));
+ if (leftWindow < 0 or rightWindow < 0)
+ util::myThrow(fmt::format("Invalid negative values for leftWindow({}) or rightWindow({})", leftWindow, rightWindow));
+
auto subModuleType = sm.str(3);
auto subModuleArguments = util::split(sm.str(4), ' ');
@@ -54,27 +57,30 @@ std::size_t RawInputModuleImpl::getInputSize()
return leftWindow + rightWindow + 1;
}
-void RawInputModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void RawInputModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
- if (leftWindow < 0 or rightWindow < 0)
- return;
-
std::string prefix = "LETTER";
-
auto & dict = getDict();
- for (auto & contextElement : context)
+
+ int insertIndex = 0;
+ for (int i = 0; i < leftWindow; i++)
{
- for (int i = 0; i < leftWindow; i++)
- if (config.hasCharacter(config.getCharacterIndex()-leftWindow+i))
- contextElement.push_back(dict.getIndexOrInsert(fmt::format("{}", config.getLetter(config.getCharacterIndex()-leftWindow+i)), prefix));
- else
- contextElement.push_back(dict.getIndexOrInsert(Dict::nullValueStr, prefix));
-
- for (int i = 0; i <= rightWindow; i++)
- if (config.hasCharacter(config.getCharacterIndex()+i))
- contextElement.push_back(dict.getIndexOrInsert(fmt::format("{}", config.getLetter(config.getCharacterIndex()+i)), prefix));
- else
- contextElement.push_back(dict.getIndexOrInsert(Dict::nullValueStr, prefix));
+ if (config.hasCharacter(config.getCharacterIndex()-leftWindow+i))
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(fmt::format("{}", config.getLetter(config.getCharacterIndex()-leftWindow+i)), prefix);
+ else
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::nullValueStr, prefix);
+
+ insertIndex++;
+ }
+
+ for (int i = 0; i <= rightWindow; i++)
+ {
+ if (config.hasCharacter(config.getCharacterIndex()+i))
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(fmt::format("{}", config.getLetter(config.getCharacterIndex()+i)), prefix);
+ else
+ context[firstInputIndex+insertIndex] = dict.getIndexOrInsert(Dict::nullValueStr, prefix);
+
+ insertIndex++;
}
}
diff --git a/torch_modules/src/SplitTransModule.cpp b/torch_modules/src/SplitTransModule.cpp
index 0c1de2e..dcb78e1 100644
--- a/torch_modules/src/SplitTransModule.cpp
+++ b/torch_modules/src/SplitTransModule.cpp
@@ -51,16 +51,15 @@ std::size_t SplitTransModuleImpl::getInputSize()
return maxNbTrans;
}
-void SplitTransModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void SplitTransModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
auto & dict = getDict();
auto & splitTransitions = config.getAppliableSplitTransitions();
- for (auto & contextElement : context)
- for (int i = 0; i < maxNbTrans; i++)
- if (i < (int)splitTransitions.size())
- contextElement.emplace_back(dict.getIndexOrInsert(splitTransitions[i]->getName(), ""));
- else
- contextElement.emplace_back(dict.getIndexOrInsert(Dict::nullValueStr, ""));
+ for (int i = 0; i < maxNbTrans; i++)
+ if (i < (int)splitTransitions.size())
+ context[firstInputIndex+i] = dict.getIndexOrInsert(splitTransitions[i]->getName(), "");
+ else
+ context[firstInputIndex+i] = dict.getIndexOrInsert(Dict::nullValueStr, "");
}
void SplitTransModuleImpl::registerEmbeddings()
diff --git a/torch_modules/src/StateNameModule.cpp b/torch_modules/src/StateNameModule.cpp
index 0c642b9..f3ac977 100644
--- a/torch_modules/src/StateNameModule.cpp
+++ b/torch_modules/src/StateNameModule.cpp
@@ -29,11 +29,10 @@ std::size_t StateNameModuleImpl::getInputSize()
return 1;
}
-void StateNameModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void StateNameModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
auto & dict = getDict();
- for (auto & contextElement : context)
- contextElement.emplace_back(dict.getIndexOrInsert(config.getState(), ""));
+ context[firstInputIndex] = dict.getIndexOrInsert(config.getState(), "");
}
void StateNameModuleImpl::registerEmbeddings()
diff --git a/torch_modules/src/UppercaseRateModule.cpp b/torch_modules/src/UppercaseRateModule.cpp
index 0452eb8..7c84615 100644
--- a/torch_modules/src/UppercaseRateModule.cpp
+++ b/torch_modules/src/UppercaseRateModule.cpp
@@ -56,7 +56,7 @@ std::size_t UppercaseRateModuleImpl::getInputSize()
return focusedBuffer.size() + focusedStack.size();
}
-void UppercaseRateModuleImpl::addToContext(std::vector<std::vector<long>> & context, const Config & config)
+void UppercaseRateModuleImpl::addToContext(torch::Tensor & context, const Config & config)
{
std::vector<long> focusedIndexes;
@@ -69,25 +69,24 @@ void UppercaseRateModuleImpl::addToContext(std::vector<std::vector<long>> & cont
else
focusedIndexes.emplace_back(-1);
- for (auto & contextElement : context)
+ int insertIndex = 0;
+ for (auto index : focusedIndexes)
{
- for (auto index : focusedIndexes)
+ double res = -1.0;
+ if (index >= 0)
{
- double res = -1.0;
- if (index >= 0)
- {
- auto word = util::splitAsUtf8(config.getAsFeature("FORM", index).get());
- int nbUpper = 0;
- for (auto & letter : word)
- if (util::isUppercase(letter))
- nbUpper++;
- if (word.size() > 0)
- res = 1.0*nbUpper/word.size();
- }
-
- contextElement.emplace_back(0);
- std::memcpy(&contextElement.back(), &res, sizeof res);
+ auto word = util::splitAsUtf8(config.getAsFeature("FORM", index).get());
+ int nbUpper = 0;
+ for (auto & letter : word)
+ if (util::isUppercase(letter))
+ nbUpper++;
+ if (word.size() > 0)
+ res = 1.0*nbUpper/word.size();
}
+
+ //TODO : Check if this works
+ context[firstInputIndex+insertIndex] = res;
+ insertIndex++;
}
}
diff --git a/trainer/include/Trainer.hpp b/trainer/include/Trainer.hpp
index dfa465e..a5088d0 100644
--- a/trainer/include/Trainer.hpp
+++ b/trainer/include/Trainer.hpp
@@ -34,7 +34,7 @@ class Trainer
int lastSavedIndex{0};
void saveIfNeeded(const std::string & state, std::filesystem::path dir, std::size_t threshold, int currentEpoch, bool dynamicOracle);
- void addContext(std::vector<std::vector<long>> & context);
+ void addContext(torch::Tensor & context);
void addClass(const LossFunction & lossFct, int nbClasses, const std::vector<long> & goldIndexes);
};
diff --git a/trainer/src/Trainer.cpp b/trainer/src/Trainer.cpp
index 56ecc44..4143b2c 100644
--- a/trainer/src/Trainer.cpp
+++ b/trainer/src/Trainer.cpp
@@ -65,7 +65,7 @@ void Trainer::extractExamples(std::vector<SubConfig> & configs, bool debug, std:
auto appliableTransitions = machine.getTransitionSet(config.getState()).getAppliableTransitions(config);
config.setAppliableTransitions(appliableTransitions);
- std::vector<std::vector<long>> context;
+ torch::Tensor context;
try
{
@@ -92,8 +92,7 @@ void Trainer::extractExamples(std::vector<SubConfig> & configs, bool debug, std:
if (dynamicOracle and util::choiceWithProbability(1.0) and config.getState() != "tokenizer" and config.getState() != "segmenter")
{
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()->forward(neuralInput, config.getState()).squeeze(0) : torch::softmax(classifier.getNN()->forward(neuralInput, config.getState()).squeeze(0), 0);
+ auto prediction = classifier.isRegression() ? classifier.getNN()->forward(context, config.getState()).squeeze(0) : torch::softmax(classifier.getNN()->forward(context, config.getState()).squeeze(0), 0);
entropy = NeuralNetworkImpl::entropy(prediction);
std::vector<int> candidates;
@@ -154,7 +153,7 @@ void Trainer::extractExamples(std::vector<SubConfig> & configs, bool debug, std:
if (!exampleIsBanned)
{
- totalNbExamples += context.size();
+ totalNbExamples += 1;
if (totalNbExamples >= (int)safetyNbExamplesMax)
util::myThrow(fmt::format("Trying to extract more examples than the limit ({})", util::int2HumanStr(safetyNbExamplesMax)));
@@ -295,12 +294,11 @@ void Trainer::Examples::saveIfNeeded(const std::string & state, std::filesystem:
classes.clear();
}
-void Trainer::Examples::addContext(std::vector<std::vector<long>> & context)
+void Trainer::Examples::addContext(torch::Tensor & context)
{
- for (auto & element : context)
- contexts.emplace_back(torch::from_blob(element.data(), {(long)element.size()}, torch::TensorOptions(torch::kLong)).clone());
+ contexts.emplace_back(context);
- currentExampleIndex += context.size();
+ currentExampleIndex += 1;
}
void Trainer::Examples::addClass(const LossFunction & lossFct, int nbClasses, const std::vector<long> & goldIndexes)
--
GitLab