Update UFSAC.py

UFSAC.py

@@ -2,6 +2,9 @@ import datasets
 import xml.etree.cElementTree as ET
 from glob import glob
 import os
+import gc
+# from memory_profiler import profile
+import objgraph
 
 _UFSAC_FILE = 'ufsac-public-2.1.tar.xz'
 
@@ -25,24 +28,30 @@ class UFSAC(datasets.GeneratorBasedBuilder):
     
     def _split_generators(self, dl_manager):
         data_dir = dl_manager.download_and_extract(_UFSAC_FILE)
-        return datasets.SplitGenerator(name = datasets.Split.TRAIN, gen_kwargs={'data_dir': data_dir})
+        return datasets.SplitGenerator(name = datasets.Split.TRAIN, gen_kwargs={'data_dir': data_dir}),
     
     def _generate_examples(self, data_dir):
         used_sents = set()
         count = 0
-        for file in glob(os.path.join(data_dir, '*.xml')):
-            tree = ET.parse(file)
-            root = tree.getroot()
-            for sent in root.iter('sentence'):
+        for file in glob(os.path.join(data_dir, 'ufsac-public-2.1/*.xml')):
+            context = ET.iterparse(file, events=('start', 'end'))
+            event, root = next(context)
+            for event, element in context:
+                if element.tag == 'paragraph':
+                    para = element
+                if element.tag != 'sentence':
+                    continue
+                if event == 'end' and element.tag == 'sentence':
+                    para.remove(element)
+                sent = element
                 words = sent.findall('word')
-                tokens = [token.attrib['surface_form'] for token in words]
-                sent_key = ''.join(tokens)
-                if sent_key not in used_sents:
-                    used_sents.add(sent_key)
-                else:
+                tokens = [token.attrib['surface_form'] if 'surface_form' in token.attrib else '_' for token in words]
+                sent_key = ''.join([token.lower() for token in tokens])
+                if sent_key in used_sents:
                     continue
-                lemmas = [token.attrib['lemma'] for token in words]
-                pos_tags = [token.attrib['pos'] for token in words]
+                used_sents.add(sent_key)
+                lemmas = [token.attrib['lemma'] if 'lemma' in token.attrib else '_' for token in words]
+                pos_tags = [token.attrib['pos'] if 'pos' in token.attrib else '_' for token in words]
                 for index, word in enumerate(words):
                     if 'wn30_key' in word.attrib:
                         senses = word.attrib['wn30_key'].split(';')
@@ -50,6 +59,7 @@ class UFSAC(datasets.GeneratorBasedBuilder):
                             'tokens': tokens,
                             'lemmas': lemmas,
                             'pos_tags': pos_tags,
-                            'target_index': index,
+                            'target_idx': index,
                             'sense_keys': senses
                         }
+                        count+=1