Improving representation speed ~5%

wani.py

@@ -881,6 +881,7 @@ class WordMsdRenderer:
         self.frequent_words = {}
         self.lemma_msd = {}
         self.lemma_features = lemma_features
+        self.memoized_msd_merges = {}
     
     def add_words(self, words):
         self.all_words.extend(words)
@@ -918,8 +919,11 @@ class WordMsdRenderer:
                     l1 if l1 != "-" else l2 for l1, l2 in zip(lf[cmsd[0]], cmsd)
                 )
         
-    @staticmethod
-    def merge_msd(common_msd, new_msd):
+    def merge_msd(self, common_msd, new_msd):
+        key = (common_msd, new_msd)
+        if key in self.memoized_msd_merges:
+            return self.memoized_msd_merges[key]
+
         def merge_letter(l1, l2):
             if l1 == "*":
                 return l2
@@ -928,7 +932,9 @@ class WordMsdRenderer:
             else:
                 return l1
 
-        return "".join(merge_letter(l1, l2) for l1, l2 in zip(common_msd, new_msd))
+        value = "".join(merge_letter(l1, l2) for l1, l2 in zip(common_msd, new_msd))
+        self.memoized_msd_merges[key] = value
+        return value
         
     def render(self, lemma, msd):
         if lemma in self.rendered_words: