Change to use ArrowBasedBuilder

README.md

@@ -127,4 +127,4 @@ size_categories:
 
 Preprocessed Wikipedia texts generated with the scripts in [singletongue/wikipedia-utils](https://github.com/singletongue/wikipedia-utils) repo.
 
-For detailed information on how the texts are processed, please refer to the repo. 
+For detailed information on how the texts are processed, please refer to the repo.

wikipedia-utils.py

@@ -1,4 +1,5 @@
 # Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+# Copyright 2023 Masatoshi Suzuki (@singletongue)
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,11 +14,11 @@
 # limitations under the License.
 """Wikipedia-Utils: Preprocessed Wikipedia Texts for NLP"""
 
-
-import json
-from typing import Dict, Iterator, List, Tuple, Union
+import io
+from typing import Iterator, List, Tuple
 
 import datasets
+import pyarrow as pa
 
 
 _DESCRIPTION = "Preprocessed Wikipedia texts generated with scripts in singletongue/wikipedia-utils repo."
@@ -40,7 +41,7 @@ _URLS = {
 _VERSION = datasets.Version("1.0.0")
 
 
-class WikipediaUtils(datasets.GeneratorBasedBuilder):
+class WikipediaUtils(datasets.ArrowBasedBuilder):
     """Wikipedia-Utils dataset."""
 
     BUILDER_CONFIGS = [datasets.BuilderConfig(name=name, version=_VERSION) for name in _URLS.keys()]
@@ -87,18 +88,36 @@ class WikipediaUtils(datasets.GeneratorBasedBuilder):
         filepath = dl_manager.download_and_extract(url)
         return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": filepath})]
 
-    def _generate_examples(self, filepath: str) -> Iterator[Tuple[int, Dict[str, Union[int, str]]]]:
+    def _generate_tables(self, filepath: str, chunksize: int = 10 << 20) -> Iterator[Tuple[int, pa.Table]]:
         if self.config.name.startswith("corpus"):
             with open(filepath) as f:
-                for id_, line in enumerate(f):
-                    line = line.rstrip("\n")
-                    yield id_, {"text": line}
-
+                batch_idx = 0
+                while True:
+                    batch = f.read(chunksize)
+                    if not batch:
+                        break
+
+                    batch += f.readline()
+                    batch = [line.rstrip("\n") for line in io.StringIO(batch).readlines()]
+                    pa_table = pa.Table.from_arrays([pa.array(batch)], names=["text"])
+
+                    yield batch_idx, pa_table
+                    batch_idx += 1
         elif self.config.name.startswith(("paragraphs", "passages")):
-            with open(filepath) as f:
-                for line in f:
-                    item = json.loads(line)
-                    yield item["id"], item
-
+            with open(filepath, "rb") as f:
+                batch_idx = 0
+                block_size = max(chunksize // 32, 16 << 10)
+                while True:
+                    batch = f.read(chunksize)
+                    if not batch:
+                        break
+
+                    batch += f.readline()
+                    pa_table = pa.json.read_json(
+                        io.BytesIO(batch), read_options=pa.json.ReadOptions(block_size=block_size)
+                    )
+
+                    yield batch_idx, pa_table
+                    batch_idx += 1
         else:
             raise ValueError("Invalid dataset config name is specified.")