Migrating PennTreebank to datapipes

torchtext/datasets/penntreebank.py

@@ -1,42 +1,53 @@
-import logging
-from torchtext.utils import download_from_url
+import os
+from typing import Union, Tuple
+
+from torchtext._internal.module_utils import is_module_available
 from torchtext.data.datasets_utils import (
-    _RawTextIterableDataset,
     _wrap_split_argument,
     _add_docstring_header,
     _create_dataset_directory,
-    _read_text_iterator,
 )
 
+if is_module_available("torchdata"):
+    from torchdata.datapipes.iter import FileOpener, HttpReader, IterableWrapper
+
 URL = {
-    'train': "https://raw.githubusercontent.com/wojzaremba/lstm/master/data/ptb.train.txt",
-    'test': "https://raw.githubusercontent.com/wojzaremba/lstm/master/data/ptb.test.txt",
-    'valid': "https://raw.githubusercontent.com/wojzaremba/lstm/master/data/ptb.valid.txt",
+    "train": "https://raw.githubusercontent.com/wojzaremba/lstm/master/data/ptb.train.txt",
+    "test": "https://raw.githubusercontent.com/wojzaremba/lstm/master/data/ptb.test.txt",
+    "valid": "https://raw.githubusercontent.com/wojzaremba/lstm/master/data/ptb.valid.txt",
 }
 
 MD5 = {
-    'train': "f26c4b92c5fdc7b3f8c7cdcb991d8420",
-    'valid': "aa0affc06ff7c36e977d7cd49e3839bf",
-    'test': "8b80168b89c18661a38ef683c0dc3721",
+    "train": "f26c4b92c5fdc7b3f8c7cdcb991d8420",
+    "valid": "aa0affc06ff7c36e977d7cd49e3839bf",
+    "test": "8b80168b89c18661a38ef683c0dc3721",
 }
 
 NUM_LINES = {
-    'train': 42068,
-    'valid': 3370,
-    'test': 3761,
+    "train": 42068,
+    "valid": 3370,
+    "test": 3761,
 }
 
 DATASET_NAME = "PennTreebank"
 
 
 @_add_docstring_header(num_lines=NUM_LINES)
 @_create_dataset_directory(dataset_name=DATASET_NAME)
-@_wrap_split_argument(('train', 'valid', 'test'))
-def PennTreebank(root, split):
-    path = download_from_url(URL[split],
-                             root=root, hash_value=MD5[split],
-                             hash_type='md5')
-    logging.info('Creating {} data'.format(split))
-    return _RawTextIterableDataset(DATASET_NAME,
-                                   NUM_LINES[split],
-                                   _read_text_iterator(path))
+@_wrap_split_argument(("train", "valid", "test"))
+def PennTreebank(root, 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[split]])
+    cache_dp = url_dp.on_disk_cache(
+        filepath_fn=lambda x: os.path.join(root, os.path.basename(x)),
+        hash_dict={os.path.join(root, os.path.basename(URL[split])): MD5[split]},
+        hash_type="md5",
+    )
+    cache_dp = HttpReader(cache_dp).end_caching(mode="w", same_filepath_fn=True)
+    data_dp = FileOpener(cache_dp, mode="r")
+    # remove single leading and trailing space from the dataset
+    return data_dp.readlines(return_path=False).map(lambda t: t.strip())