Support streaming (#1)

- Use iter_files (d1d4571712a169dce2d2f3f211e6bbf323d1a121)
- Optimize local parse_brat_file (90bf9d3ea9f44b6a375e95186abe85aac9c6672c)


Co-authored-by: Albert Villanova <[email protected]>

chia.py

@@ -232,20 +232,21 @@ class ChiaDataset(datasets.GeneratorBasedBuilder):
             url_key += "_wo_scope"
 
         urls = _URLS[url_key]
-        data_dir = Path(dl_manager.download_and_extract(urls))
+        data_dir = dl_manager.download_and_extract(urls)
 
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
-                gen_kwargs={"data_dir": data_dir},
+                gen_kwargs={"file_paths": dl_manager.iter_files(data_dir)},
             )
         ]
 
-    def _generate_examples(self, data_dir: Path) -> Iterator[Tuple[str, Dict]]:
+    def _generate_examples(self, file_paths: Iterator[str]) -> Iterator[Tuple[str, Dict]]:
         if self.config.schema == "source":
             fix_offsets = "fixed" in self.config.subset_id
 
-            for file in data_dir.iterdir():
+            for file_path in file_paths:
+                file = Path(file_path)
                 if not file.name.endswith(".txt"):
                     continue
 
@@ -256,7 +257,8 @@ class ChiaDataset(datasets.GeneratorBasedBuilder):
                 yield source_example["id"], source_example
 
         elif self.config.schema == "bigbio_kb":
-            for file in data_dir.iterdir():
+            for file_path in file_paths:
+                file = Path(file_path)
                 if not file.name.endswith(".txt"):
                     continue
 
@@ -518,9 +520,11 @@ def parse_brat_file(txt_file: Path, annotation_file_suffixes: List[str] = None)
     ann_lines = []
     for suffix in annotation_file_suffixes:
         annotation_file = txt_file.with_suffix(suffix)
-        if annotation_file.exists():
+        try:
             with annotation_file.open() as f:
                 ann_lines.extend(f.readlines())
+        except Exception:
+            continue
 
     example["text_bound_annotations"] = []
     example["events"] = []