Migrate IMDB to datapipes

torchtext/datasets/imdb.py

@@ -1,10 +1,15 @@
-from torchtext.utils import download_from_url, extract_archive
-from torchtext.data.datasets_utils import _RawTextIterableDataset
-from torchtext.data.datasets_utils import _wrap_split_argument
+import os
+from pathlib import Path
+from typing import Union, Tuple
+
+from torchtext._internal.module_utils import is_module_available
 from torchtext.data.datasets_utils import _add_docstring_header
 from torchtext.data.datasets_utils import _create_dataset_directory
-import io
-from pathlib import Path
+from torchtext.data.datasets_utils import _wrap_split_argument
+
+if is_module_available("torchdata"):
+    from torchdata.datapipes.iter import FileOpener, IterableWrapper, HttpReader
+
 
 URL = 'http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz'
 
@@ -23,16 +28,41 @@
 @_add_docstring_header(num_lines=NUM_LINES, num_classes=2)
 @_create_dataset_directory(dataset_name=DATASET_NAME)
 @_wrap_split_argument(('train', 'test'))
-def IMDB(root, split):
-    def generate_imdb_data(key, extracted_files):
-        for fname in extracted_files:
-            *_, split, label, file = Path(fname).parts
-
-            if key == split and (label in ['pos', 'neg']):
-                with io.open(fname, encoding="utf8") as f:
-                    yield label, f.read()
-    dataset_tar = download_from_url(URL, root=root,
-                                    hash_value=MD5, hash_type='md5')
-    extracted_files = extract_archive(dataset_tar)
-    iterator = generate_imdb_data(split, extracted_files)
-    return _RawTextIterableDataset(DATASET_NAME, NUM_LINES[split], iterator)
+def IMDB(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_compressed_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_compressed_dp = HttpReader(cache_compressed_dp).end_caching(mode="wb", same_filepath_fn=True)
+
+    labels = {"neg", "pos"}
+    decompressed_folder = "aclImdb_v1"
+    cache_decompressed_dp = cache_compressed_dp.on_disk_cache(
+        filepath_fn=lambda x: [os.path.join(root, decompressed_folder, split, label) for label in labels]
+    )
+    cache_decompressed_dp = FileOpener(cache_decompressed_dp, mode="b")
+    cache_decompressed_dp = cache_decompressed_dp.read_from_tar()
+
+    def filter_imdb_data(key, fname):
+        # eg. fname = "aclImdb/train/neg/12416_3.txt"
+        *_, split, label, file = Path(fname).parts
+        return key == split and label in labels
+
+    cache_decompressed_dp = cache_decompressed_dp.filter(lambda t: filter_imdb_data(split, t[0]))
+
+    # eg. "aclImdb/train/neg/12416_3.txt" -> "neg"
+    cache_decompressed_dp = cache_decompressed_dp.map(lambda t: (Path(t[0]).parts[-2], t[1]))
+    cache_decompressed_dp = cache_decompressed_dp.readlines(decode=True)
+    cache_decompressed_dp = cache_decompressed_dp.lines_to_paragraphs()  # group by label in cache file
+    cache_decompressed_dp = cache_decompressed_dp.end_caching(
+        mode="wt", filepath_fn=lambda x: os.path.join(root, decompressed_folder, split, x)
+    )
+
+    data_dp = FileOpener(cache_decompressed_dp, mode="t")
+    # get label from cache file, eg. "aclImdb_v1/train/neg" -> "neg"
+    return data_dp.readlines().map(lambda t: (Path(t[0]).parts[-1], t[1]))