From 95131d6d8186067630bb3457395ad04071964833 Mon Sep 17 00:00:00 2001
From: Franck Dary <franck.dary@lis-lab.fr>
Date: Wed, 10 Jun 2020 13:36:31 +0200
Subject: [PATCH] Added DistanceModule
---
reading_machine/include/Config.hpp | 1 +
reading_machine/src/Config.cpp | 26 +++++
torch_modules/include/DistanceModule.hpp | 34 +++++++
torch_modules/include/ModularNetwork.hpp | 1 +
torch_modules/src/DistanceModule.cpp | 115 +++++++++++++++++++++++
torch_modules/src/ModularNetwork.cpp | 2 +
6 files changed, 179 insertions(+)
create mode 100644 torch_modules/include/DistanceModule.hpp
create mode 100644 torch_modules/src/DistanceModule.cpp
diff --git a/reading_machine/include/Config.hpp b/reading_machine/include/Config.hpp
index faa594e..71fb1ee 100644
--- a/reading_machine/include/Config.hpp
+++ b/reading_machine/include/Config.hpp
@@ -98,6 +98,7 @@ class Config
std::size_t & getStackRef(int relativeIndex);
long getRelativeWordIndex(int relativeIndex) const;
+ long getRelativeDistance(int fromIndex, int toIndex) const;
public :
diff --git a/reading_machine/src/Config.cpp b/reading_machine/src/Config.cpp
index 4e320b9..c967a4f 100644
--- a/reading_machine/src/Config.cpp
+++ b/reading_machine/src/Config.cpp
@@ -675,6 +675,32 @@ long Config::getRelativeWordIndex(int relativeIndex) const
return -1;
}
+long Config::getRelativeDistance(int fromIndex, int toIndex) const
+{
+ if (toIndex < fromIndex)
+ {
+ for (int index = fromIndex, counter = 0; has(0,index,0); --index)
+ if (!isCommentPredicted(index))
+ {
+ if (index == toIndex)
+ return counter;
+ --counter;
+ }
+ }
+ else
+ {
+ for (int index = fromIndex, counter = 0; has(0,index,0); ++index)
+ if (!isCommentPredicted(index))
+ {
+ if (index == toIndex)
+ return counter;
+ ++counter;
+ }
+ }
+
+ return 0;
+}
+
long Config::getRelativeWordIndex(Object object, int relativeIndex) const
{
if (object == Object::Buffer)
diff --git a/torch_modules/include/DistanceModule.hpp b/torch_modules/include/DistanceModule.hpp
new file mode 100644
index 0000000..b6e22d8
--- /dev/null
+++ b/torch_modules/include/DistanceModule.hpp
@@ -0,0 +1,34 @@
+#ifndef DISTANCEMODULE__H
+#define DISTANCEMODULE__H
+
+#include <torch/torch.h>
+#include "Submodule.hpp"
+#include "MyModule.hpp"
+#include "LSTM.hpp"
+#include "GRU.hpp"
+#include "Concat.hpp"
+
+class DistanceModuleImpl : public Submodule
+{
+ private :
+
+ torch::nn::Embedding wordEmbeddings{nullptr};
+ std::shared_ptr<MyModule> myModule{nullptr};
+ std::vector<int> fromBuffer, fromStack;
+ std::vector<int> toBuffer, toStack;
+ int threshold;
+ int inSize;
+
+ public :
+
+ DistanceModuleImpl(std::string name, const std::string & definition);
+ 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 registerEmbeddings(std::filesystem::path pretrained) override;
+};
+TORCH_MODULE(DistanceModule);
+
+#endif
+
diff --git a/torch_modules/include/ModularNetwork.hpp b/torch_modules/include/ModularNetwork.hpp
index a6a6c3e..7e98302 100644
--- a/torch_modules/include/ModularNetwork.hpp
+++ b/torch_modules/include/ModularNetwork.hpp
@@ -12,6 +12,7 @@
#include "UppercaseRateModule.hpp"
#include "NumericColumnModule.hpp"
#include "HistoryModule.hpp"
+#include "DistanceModule.hpp"
#include "MLP.hpp"
class ModularNetworkImpl : public NeuralNetworkImpl
diff --git a/torch_modules/src/DistanceModule.cpp b/torch_modules/src/DistanceModule.cpp
new file mode 100644
index 0000000..50deea0
--- /dev/null
+++ b/torch_modules/src/DistanceModule.cpp
@@ -0,0 +1,115 @@
+#include "DistanceModule.hpp"
+
+DistanceModuleImpl::DistanceModuleImpl(std::string name, const std::string & definition)
+{
+ setName(name);
+ std::regex regex("(?:(?:\\s|\\t)*)FromBuffer\\{(.*)\\}(?:(?:\\s|\\t)*)FromStack\\{(.*)\\}(?:(?:\\s|\\t)*)ToBuffer\\{(.*)\\}(?:(?:\\s|\\t)*)ToStack\\{(.*)\\}(?:(?:\\s|\\t)*)Threshold\\{(.*)\\}(?:(?:\\s|\\t)*)(\\S+)\\{(.*)\\}(?:(?:\\s|\\t)*)In\\{(.*)\\}(?:(?:\\s|\\t)*)Out\\{(.*)\\}(?:(?:\\s|\\t)*)");
+ if (!util::doIfNameMatch(regex, definition, [this,&definition](auto sm)
+ {
+ try
+ {
+ for (auto & index : util::split(sm.str(1), ' '))
+ fromBuffer.emplace_back(std::stoi(index));
+
+ for (auto & index : util::split(sm.str(2), ' '))
+ fromStack.emplace_back(std::stoi(index));
+
+ for (auto & index : util::split(sm.str(3), ' '))
+ toBuffer.emplace_back(std::stoi(index));
+
+ for (auto & index : util::split(sm.str(4), ' '))
+ toStack.emplace_back(std::stoi(index));
+
+ threshold = std::stoi(sm.str(5));
+
+ auto subModuleType = sm.str(6);
+ auto subModuleArguments = util::split(sm.str(7), ' ');
+
+ auto options = MyModule::ModuleOptions(true)
+ .bidirectional(std::stoi(subModuleArguments[0]))
+ .num_layers(std::stoi(subModuleArguments[1]))
+ .dropout(std::stof(subModuleArguments[2]))
+ .complete(std::stoi(subModuleArguments[3]));
+
+ inSize = std::stoi(sm.str(8));
+ int outSize = std::stoi(sm.str(9));
+
+ if (subModuleType == "LSTM")
+ myModule = register_module("myModule", LSTM(inSize, outSize, options));
+ else if (subModuleType == "GRU")
+ myModule = register_module("myModule", GRU(inSize, outSize, options));
+ else if (subModuleType == "Concat")
+ myModule = register_module("myModule", Concat(inSize));
+ else
+ util::myThrow(fmt::format("unknown sumodule type '{}'", subModuleType));
+
+ } catch (std::exception & e) {util::myThrow(fmt::format("{} in '{}'",e.what(),definition));}
+ }))
+ util::myThrow(fmt::format("invalid definition '{}'", definition));
+}
+
+torch::Tensor DistanceModuleImpl::forward(torch::Tensor input)
+{
+ return myModule->forward(wordEmbeddings(input.narrow(1, firstInputIndex, getInputSize())));
+}
+
+std::size_t DistanceModuleImpl::getOutputSize()
+{
+ return myModule->getOutputSize(getInputSize());
+}
+
+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)
+{
+ auto & dict = getDict();
+ std::vector<long> fromIndexes, toIndexes;
+
+ for (int index : fromBuffer)
+ fromIndexes.emplace_back(config.getRelativeWordIndex(index));
+
+ for (int index : fromStack)
+ if (config.hasStack(index))
+ fromIndexes.emplace_back(config.getStack(index));
+ else
+ fromIndexes.emplace_back(-1);
+
+ for (int index : toBuffer)
+ toIndexes.emplace_back(config.getRelativeWordIndex(index));
+
+ for (int index : toStack)
+ if (config.hasStack(index))
+ toIndexes.emplace_back(config.getStack(index));
+ else
+ toIndexes.emplace_back(-1);
+
+ for (auto & contextElement : context)
+ {
+ for (auto from : fromIndexes)
+ for (auto to : toIndexes)
+ {
+ if (from == -1 or to == -1)
+ {
+ contextElement.emplace_back(dict.getIndexOrInsert(Dict::nullValueStr));
+ continue;
+ }
+
+ long dist = std::abs(config.getRelativeDistance(from, to));
+
+ if (dist <= threshold)
+ contextElement.emplace_back(dict.getIndexOrInsert(fmt::format("{}", dist)));
+ else
+ contextElement.emplace_back(dict.getIndexOrInsert(Dict::unknownValueStr));
+ }
+ }
+}
+
+void DistanceModuleImpl::registerEmbeddings(std::filesystem::path path)
+{
+ wordEmbeddings = register_module("embeddings", torch::nn::Embedding(torch::nn::EmbeddingOptions(getDict().size(), inSize)));
+ loadPretrainedW2vEmbeddings(wordEmbeddings, path);
+}
+
diff --git a/torch_modules/src/ModularNetwork.cpp b/torch_modules/src/ModularNetwork.cpp
index f9707a1..22cdb3a 100644
--- a/torch_modules/src/ModularNetwork.cpp
+++ b/torch_modules/src/ModularNetwork.cpp
@@ -45,6 +45,8 @@ ModularNetworkImpl::ModularNetworkImpl(std::string name, std::map<std::string,st
modules.emplace_back(register_module(name, SplitTransModule(nameH, Config::maxNbAppliableSplitTransitions, splited.second)));
else if (splited.first == "AppliableTrans")
modules.emplace_back(register_module(name, AppliableTransModule(nameH, maxNbOutputs)));
+ else if (splited.first == "Distance")
+ modules.emplace_back(register_module(name, DistanceModule(nameH, splited.second)));
else if (splited.first == "DepthLayerTree")
modules.emplace_back(register_module(name, DepthLayerTreeEmbeddingModule(nameH, splited.second)));
else if (splited.first == "MLP")
--
GitLab