From ce0503c84319a5918dee1dba2e6949d68d109397 Mon Sep 17 00:00:00 2001
From: Franck Dary <franck.dary@lis-lab.fr>
Date: Mon, 20 Apr 2020 10:28:42 +0200
Subject: [PATCH] Do not split for unknown tokens but replace instead

---
 torch_modules/src/ContextLSTM.cpp | 13 ++++---------
 1 file changed, 4 insertions(+), 9 deletions(-)

diff --git a/torch_modules/src/ContextLSTM.cpp b/torch_modules/src/ContextLSTM.cpp
index 5da02e7..d247788 100644
--- a/torch_modules/src/ContextLSTM.cpp
+++ b/torch_modules/src/ContextLSTM.cpp
@@ -42,15 +42,10 @@ void ContextLSTMImpl::addToContext(std::vector<std::vector<long>> & context, Dic
         for (auto & contextElement : context)
           contextElement.push_back(dictIndex);
 
-
-        if (splitUnknown)
-          for (auto & targetCol : unknownValueColumns)
-            if (col == targetCol)
-              if (dict.getNbOccs(dictIndex) <= unknownValueThreshold)
-              {
-                context.emplace_back(context.back());
-                context.back().back() = dict.getIndexOrInsert(Dict::unknownValueStr);
-              }
+        for (auto & targetCol : unknownValueColumns)
+          if (col == targetCol)
+            if (dict.getNbOccs(dictIndex) <= unknownValueThreshold)
+              context.back().back() = dict.getIndexOrInsert(Dict::unknownValueStr);
       }
 }
 
-- 
GitLab