Commit
•
e1ebc67
1
Parent(s):
6f885f7
Update wmt_utils with Yakut
Browse files- wmt_utils.py +23 -3
wmt_utils.py
CHANGED
@@ -409,6 +409,13 @@ _TRAIN_SUBSETS = [
|
|
409 |
url="https://huggingface.co/datasets/wmt/wikititles/resolve/main/v1/wikititles-v1.{src}-en.tsv.gz",
|
410 |
path="",
|
411 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
412 |
SubDataset(
|
413 |
name="yandexcorpus",
|
414 |
target="en",
|
@@ -799,6 +806,10 @@ class Wmt(datasets.GeneratorBasedBuilder):
|
|
799 |
sub_generator = _parse_czeng
|
800 |
elif ss_name == "hindencorp_01":
|
801 |
sub_generator = _parse_hindencorp
|
|
|
|
|
|
|
|
|
802 |
elif len(files) == 2:
|
803 |
if ss_name.endswith("_frde"):
|
804 |
sub_generator = _parse_frde_bitext
|
@@ -941,7 +952,7 @@ def _parse_tmx(path):
|
|
941 |
elem.clear()
|
942 |
|
943 |
|
944 |
-
def _parse_tsv(path, filename, language_pair=None):
|
945 |
"""Generates examples from TSV file."""
|
946 |
if language_pair is None:
|
947 |
lang_match = re.match(r".*\.([a-z][a-z])-([a-z][a-z])\.tsv", filename)
|
@@ -950,13 +961,15 @@ def _parse_tsv(path, filename, language_pair=None):
|
|
950 |
else:
|
951 |
l1, l2 = language_pair
|
952 |
with open(path, encoding="utf-8") as f:
|
953 |
-
for
|
|
|
|
|
954 |
cols = line.split("\t")
|
955 |
if len(cols) != 2:
|
956 |
logger.warning("Skipping line %d in TSV (%s) with %d != 2 columns.", j, path, len(cols))
|
957 |
continue
|
958 |
s1, s2 = cols
|
959 |
-
yield
|
960 |
|
961 |
|
962 |
def _parse_wikiheadlines(path):
|
@@ -1007,3 +1020,10 @@ def _parse_hindencorp(path):
|
|
1007 |
logger.warning("Skipping invalid HindEnCorp line: %s", line)
|
1008 |
continue
|
1009 |
yield line_id, {"translation": {"en": split_line[3].strip(), "hi": split_line[4].strip()}}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
409 |
url="https://huggingface.co/datasets/wmt/wikititles/resolve/main/v1/wikititles-v1.{src}-en.tsv.gz",
|
410 |
path="",
|
411 |
),
|
412 |
+
SubDataset(
|
413 |
+
name="yakut",
|
414 |
+
target="ru",
|
415 |
+
sources={"sah"},
|
416 |
+
url="https://huggingface.co/datasets/wmt/yakut/resolve/main/data/yakut.zip",
|
417 |
+
path="yakut/sah-ru.parallel.uniq.tsv",
|
418 |
+
),
|
419 |
SubDataset(
|
420 |
name="yandexcorpus",
|
421 |
target="en",
|
|
|
806 |
sub_generator = _parse_czeng
|
807 |
elif ss_name == "hindencorp_01":
|
808 |
sub_generator = _parse_hindencorp
|
809 |
+
elif ss_name == "yakut":
|
810 |
+
sub_generator, sub_generator_args = YakutParser.create_generator(
|
811 |
+
sub_generator_args=sub_generator_args, config=self.config
|
812 |
+
)
|
813 |
elif len(files) == 2:
|
814 |
if ss_name.endswith("_frde"):
|
815 |
sub_generator = _parse_frde_bitext
|
|
|
952 |
elem.clear()
|
953 |
|
954 |
|
955 |
+
def _parse_tsv(path, filename=None, language_pair=None, skiprows=None):
|
956 |
"""Generates examples from TSV file."""
|
957 |
if language_pair is None:
|
958 |
lang_match = re.match(r".*\.([a-z][a-z])-([a-z][a-z])\.tsv", filename)
|
|
|
961 |
else:
|
962 |
l1, l2 = language_pair
|
963 |
with open(path, encoding="utf-8") as f:
|
964 |
+
for key, line in enumerate(f):
|
965 |
+
if skiprows and key < skiprows:
|
966 |
+
continue
|
967 |
cols = line.split("\t")
|
968 |
if len(cols) != 2:
|
969 |
logger.warning("Skipping line %d in TSV (%s) with %d != 2 columns.", j, path, len(cols))
|
970 |
continue
|
971 |
s1, s2 = cols
|
972 |
+
yield key, {l1: s1.strip(), l2: s2.strip()}
|
973 |
|
974 |
|
975 |
def _parse_wikiheadlines(path):
|
|
|
1020 |
logger.warning("Skipping invalid HindEnCorp line: %s", line)
|
1021 |
continue
|
1022 |
yield line_id, {"translation": {"en": split_line[3].strip(), "hi": split_line[4].strip()}}
|
1023 |
+
|
1024 |
+
|
1025 |
+
class YakutParser:
|
1026 |
+
@staticmethod
|
1027 |
+
def create_generator(sub_generator_args=None, config=None):
|
1028 |
+
sub_generator = functools.partial(_parse_tsv, language_pair=config.language_pair, skiprows=1)
|
1029 |
+
return sub_generator, sub_generator_args
|