From 50f6f7e77beeadb0768258eb666658c0398fb2eb Mon Sep 17 00:00:00 2001 From: Elijah Rippeth Date: Wed, 12 Jan 2022 07:35:28 -0500 Subject: [PATCH 1/2] add initial pass at migrating YelpReviewFull to datapipes. --- torchtext/datasets/yelpreviewfull.py | 45 ++++++++++++++++++---------- 1 file changed, 29 insertions(+), 16 deletions(-) diff --git a/torchtext/datasets/yelpreviewfull.py b/torchtext/datasets/yelpreviewfull.py index 82db628b6f..9536e25021 100644 --- a/torchtext/datasets/yelpreviewfull.py +++ b/torchtext/datasets/yelpreviewfull.py @@ -1,15 +1,15 @@ -from torchtext.utils import ( - download_from_url, - extract_archive, -) +from torchtext._internal.module_utils import is_module_available +from typing import Union, Tuple + +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, - _find_match, _create_dataset_directory, - _create_data_from_csv, ) + import os URL = 'https://drive.google.com/uc?export=download&id=0Bz8a_Dbh9QhbZlU4dXhHTFhZQU0' @@ -25,16 +25,29 @@ DATASET_NAME = "YelpReviewFull" +_EXTRACTED_FILES = { + 'train': os.path.join('yelp_review_full_csv', 'train.csv'), + 'test': os.path.join('yelp_review_full_csv', 'test.csv'), +} @_add_docstring_header(num_lines=NUM_LINES, num_classes=5) @_create_dataset_directory(dataset_name=DATASET_NAME) @_wrap_split_argument(('train', 'test')) -def YelpReviewFull(root, split): - dataset_tar = download_from_url(URL, root=root, - path=os.path.join(root, _PATH), - hash_value=MD5, hash_type='md5') - extracted_files = extract_archive(dataset_tar) - - path = _find_match(split + '.csv', extracted_files) - return _RawTextIterableDataset(DATASET_NAME, NUM_LINES[split], - _create_data_from_csv(path)) +def YelpReviewFull(root: str, split: Union[Tuple[str], str]): + if not is_module_available("torchdata"): + raise ModuleNotFoundError("Package `torchdata` not found. Please install following instructions at `https://github.com/pytorch/data`") + + 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]), " ".join(t[1:]))) From a85af53c55226629b81a3ea13d78899ec8f532ee Mon Sep 17 00:00:00 2001 From: Elijah Rippeth Date: Wed, 12 Jan 2022 08:11:19 -0500 Subject: [PATCH 2/2] fix flake. --- torchtext/datasets/yelpreviewfull.py | 1 + 1 file changed, 1 insertion(+) diff --git a/torchtext/datasets/yelpreviewfull.py b/torchtext/datasets/yelpreviewfull.py index 9536e25021..270b627845 100644 --- a/torchtext/datasets/yelpreviewfull.py +++ b/torchtext/datasets/yelpreviewfull.py @@ -30,6 +30,7 @@ 'test': os.path.join('yelp_review_full_csv', 'test.csv'), } + @_add_docstring_header(num_lines=NUM_LINES, num_classes=5) @_create_dataset_directory(dataset_name=DATASET_NAME) @_wrap_split_argument(('train', 'test'))