From 675d8f42c8ae6e201aa031517c40e2f17684ef20 Mon Sep 17 00:00:00 2001
From: Franck Dary <franck.dary@lis-lab.fr>
Date: Thu, 30 Jul 2020 21:40:34 +0200
Subject: [PATCH] allow multiple pretrained embeddings file for ContextModule
---
torch_modules/include/ContextModule.hpp | 2 +-
torch_modules/src/ContextModule.cpp | 14 +++++++++-----
2 files changed, 10 insertions(+), 6 deletions(-)
diff --git a/torch_modules/include/ContextModule.hpp b/torch_modules/include/ContextModule.hpp
index 508ba71..fc24680 100644
--- a/torch_modules/include/ContextModule.hpp
+++ b/torch_modules/include/ContextModule.hpp
@@ -21,7 +21,7 @@ class ContextModuleImpl : public Submodule
std::vector<std::tuple<Config::Object, int, std::optional<int>>> targets;
int inSize;
std::filesystem::path path;
- std::filesystem::path w2vFile;
+ std::filesystem::path w2vFiles;
public :
diff --git a/torch_modules/src/ContextModule.cpp b/torch_modules/src/ContextModule.cpp
index 9016938..66a6728 100644
--- a/torch_modules/src/ContextModule.cpp
+++ b/torch_modules/src/ContextModule.cpp
@@ -48,11 +48,13 @@ ContextModuleImpl::ContextModuleImpl(std::string name, const std::string & defin
else
util::myThrow(fmt::format("unknown sumodule type '{}'", subModuleType));
- w2vFile = sm.str(7);
+ w2vFiles = sm.str(7);
- if (!w2vFile.empty())
+ if (!w2vFiles.empty())
{
- getDict().loadWord2Vec(this->path / w2vFile);
+ auto pathes = util::split(w2vFiles.string(), ' ');
+ for (auto & p : pathes)
+ getDict().loadWord2Vec(this->path / p);
getDict().setState(Dict::State::Closed);
dictSetPretrained(true);
}
@@ -138,7 +140,7 @@ void ContextModuleImpl::addToContext(std::vector<std::vector<long>> & context, c
else
{
std::string featureValue = functions[colIndex](config.getAsFeature(col, index));
- if (w2vFile.empty())
+ if (w2vFiles.empty())
featureValue = fmt::format("{}({})", col, featureValue);
dictIndex = dict.getIndexOrInsert(featureValue);
}
@@ -161,6 +163,8 @@ torch::Tensor ContextModuleImpl::forward(torch::Tensor input)
void ContextModuleImpl::registerEmbeddings()
{
wordEmbeddings = register_module("embeddings", torch::nn::Embedding(torch::nn::EmbeddingOptions(getDict().size(), inSize)));
- loadPretrainedW2vEmbeddings(wordEmbeddings, w2vFile.empty() ? "" : path / w2vFile);
+ auto pathes = util::split(w2vFiles.string(), ' ');
+ for (auto & p : pathes)
+ loadPretrainedW2vEmbeddings(wordEmbeddings, path / p);
}
--
GitLab