add initial pass at migrating Amazon Review Full to datapipes.

Author: erip

torchtext/datasets/amazonreviewfull.py

@@ -1,13 +1,15 @@
+from torchtext._internal.module_utils import is_module_available
+
+if is_module_available("torchdata"):
+    from torchdata.datapipes.iter import FileOpener, GDriveReader, IterableWrapper
+
 from torchtext.data.datasets_utils import (
-    _RawTextIterableDataset,
     _wrap_split_argument,
     _add_docstring_header,
-    _download_extract_validate,
     _create_dataset_directory,
-    _create_data_from_csv,
 )
+
 import os
-import logging
 
 URL = 'https://drive.google.com/uc?export=download&id=0Bz8a_Dbh9QhbZVhsUnRWRDhETzA'
 
@@ -35,10 +37,19 @@
 
 @_add_docstring_header(num_lines=NUM_LINES, num_classes=5)
 @_create_dataset_directory(dataset_name=DATASET_NAME)
-@_wrap_split_argument(('train', 'test'))
+@_wrap_split_argument(("train", "test"))
 def AmazonReviewFull(root, split):
-    path = _download_extract_validate(root, URL, MD5, os.path.join(root, _PATH), os.path.join(root, _EXTRACTED_FILES[split]),
-                                      _EXTRACTED_FILES_MD5[split], hash_type="md5")
-    logging.info('Creating {} data'.format(split))
-    return _RawTextIterableDataset(DATASET_NAME, NUM_LINES[split],
-                                   _create_data_from_csv(path))
+    url_dp = IterableWrapper([URL])
+
+    cache_dp = url_dp.on_disk_cache(
+        filepath_fn=lambda x: os.path.join(root, _PATH),
+        hash_dict={os.path.join(root, _PATH): MD5}, hash_type="md5"
+    )
+    cache_dp = GDriveReader(cache_dp).end_caching(mode="wb", same_filepath_fn=True)
+    cache_dp = FileOpener(cache_dp, mode="b")
+
+    extracted_files = cache_dp.read_from_tar()
+
+    filter_extracted_files = extracted_files.filter(lambda x: _EXTRACTED_FILES[split] in x[0])
+
+    return filter_extracted_files.parse_csv().map(fn=lambda t: (int(t[0]), t[1:]))