Datasets:
Introduce stratified folds
Browse files
chall.py
CHANGED
@@ -19,7 +19,6 @@ logger = logging.get_logger(__name__)
|
|
19 |
|
20 |
|
21 |
class ChallConfig(BuilderConfig):
|
22 |
-
|
23 |
split_segments: bool = False
|
24 |
|
25 |
# settings that can only be used together with split_segments
|
@@ -33,7 +32,11 @@ class ChallConfig(BuilderConfig):
|
|
33 |
allowed_chars: set
|
34 |
special_terms_mapping: dict
|
35 |
|
|
|
|
|
|
|
36 |
def __init__(self, **kwargs):
|
|
|
37 |
self.split_segments = kwargs.pop("split_segments", False)
|
38 |
self.remove_trailing_pauses = kwargs.pop("remove_trailing_pauses", False)
|
39 |
|
@@ -50,11 +53,13 @@ class ChallConfig(BuilderConfig):
|
|
50 |
else:
|
51 |
self.allowed_chars: set = set(string.ascii_lowercase + string.ascii_uppercase + " ÄÖÜäöü'")
|
52 |
|
|
|
|
|
|
|
53 |
super(ChallConfig, self).__init__(**kwargs)
|
54 |
|
55 |
|
56 |
class Chall(GeneratorBasedBuilder):
|
57 |
-
|
58 |
VERSION = Version("1.0.0")
|
59 |
|
60 |
BUILDER_CONFIG_CLASS = ChallConfig
|
@@ -84,6 +89,14 @@ class Chall(GeneratorBasedBuilder):
|
|
84 |
remove_trailing_pauses=True,
|
85 |
lowercase=True,
|
86 |
num_to_words=True,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
87 |
description="Settings used for the paper."
|
88 |
)
|
89 |
]
|
@@ -220,29 +233,47 @@ class Chall(GeneratorBasedBuilder):
|
|
220 |
f"that includes files unzipped from the chall zip. Manual download instructions: {self.manual_download_instructions}"
|
221 |
)
|
222 |
|
223 |
-
|
224 |
-
|
225 |
-
|
226 |
-
|
227 |
-
|
228 |
-
|
229 |
-
|
230 |
-
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
|
236 |
-
|
237 |
-
|
238 |
-
|
239 |
-
|
240 |
-
|
241 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
242 |
"""
|
243 |
This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
|
244 |
:param filepath: The path where the data is located.
|
245 |
:param metafile: The metafile describing the chall data
|
|
|
|
|
246 |
:return:
|
247 |
"""
|
248 |
|
@@ -251,6 +282,11 @@ class Chall(GeneratorBasedBuilder):
|
|
251 |
with open(metafile, 'r') as file:
|
252 |
metadata = json.load(file)
|
253 |
for row in metadata["data"]:
|
|
|
|
|
|
|
|
|
|
|
254 |
# load transcript
|
255 |
transcript_file = os.path.join(filepath, row["transcript_file"])
|
256 |
with open(transcript_file, 'r') as transcript:
|
@@ -386,14 +422,14 @@ class Chall(GeneratorBasedBuilder):
|
|
386 |
|
387 |
split_segments = []
|
388 |
for segment in segments:
|
389 |
-
if any(w["end"]-w["start"] >= self.config.max_pause_length and w["text"].strip() == "(...)" for w in segment["words"]):
|
390 |
start_i = 0
|
391 |
for i, word in enumerate(segment["words"]):
|
392 |
w_duration = word["end"] - word["start"]
|
393 |
if w_duration >= self.config.max_pause_length and word["text"].strip() == "(...)":
|
394 |
if len(segment["words"][start_i:i]) > 0:
|
395 |
split_segments.append({"speaker": segment["speaker"], "words": segment["words"][start_i:i]})
|
396 |
-
start_i = i+1
|
397 |
|
398 |
if len(segment["words"][start_i:]) > 0:
|
399 |
split_segments.append({"speaker": segment["speaker"], "words": segment["words"][start_i:]})
|
|
|
19 |
|
20 |
|
21 |
class ChallConfig(BuilderConfig):
|
|
|
22 |
split_segments: bool = False
|
23 |
|
24 |
# settings that can only be used together with split_segments
|
|
|
32 |
allowed_chars: set
|
33 |
special_terms_mapping: dict
|
34 |
|
35 |
+
stratify_column: Union[None, str]
|
36 |
+
folds: Union[None, Dict[str, List]]
|
37 |
+
|
38 |
def __init__(self, **kwargs):
|
39 |
+
|
40 |
self.split_segments = kwargs.pop("split_segments", False)
|
41 |
self.remove_trailing_pauses = kwargs.pop("remove_trailing_pauses", False)
|
42 |
|
|
|
53 |
else:
|
54 |
self.allowed_chars: set = set(string.ascii_lowercase + string.ascii_uppercase + " ÄÖÜäöü'")
|
55 |
|
56 |
+
self.stratify_column = kwargs.pop("stratify_column", None)
|
57 |
+
self.folds = kwargs.pop("folds", None)
|
58 |
+
|
59 |
super(ChallConfig, self).__init__(**kwargs)
|
60 |
|
61 |
|
62 |
class Chall(GeneratorBasedBuilder):
|
|
|
63 |
VERSION = Version("1.0.0")
|
64 |
|
65 |
BUILDER_CONFIG_CLASS = ChallConfig
|
|
|
89 |
remove_trailing_pauses=True,
|
90 |
lowercase=True,
|
91 |
num_to_words=True,
|
92 |
+
stratify_column="intervention",
|
93 |
+
folds={
|
94 |
+
"fold0": ["17", "15", "1"],
|
95 |
+
"fold1": ["13", "7", "10"],
|
96 |
+
"fold2": ["4", "8", "6", "14"],
|
97 |
+
"fold3": ["12", "16", "5", "19"],
|
98 |
+
"fold4": ["9", "2", "3", "18", "11"]
|
99 |
+
},
|
100 |
description="Settings used for the paper."
|
101 |
)
|
102 |
]
|
|
|
233 |
f"that includes files unzipped from the chall zip. Manual download instructions: {self.manual_download_instructions}"
|
234 |
)
|
235 |
|
236 |
+
# kFold Strategy
|
237 |
+
if self.config.folds and self.config.stratify_column:
|
238 |
+
return [
|
239 |
+
SplitGenerator(
|
240 |
+
name=fold_name,
|
241 |
+
gen_kwargs={
|
242 |
+
"filepath": os.path.join(data_dir, "data"),
|
243 |
+
"metafile": os.path.join(data_dir, _METAFILE),
|
244 |
+
"stratify_column": self.config.stratify_column,
|
245 |
+
"fold": fold
|
246 |
+
},
|
247 |
+
)
|
248 |
+
for (fold_name, fold) in self.config.folds.items()]
|
249 |
+
|
250 |
+
# Train Only Strategy
|
251 |
+
else:
|
252 |
+
return [
|
253 |
+
SplitGenerator(
|
254 |
+
name=Split.TRAIN,
|
255 |
+
gen_kwargs={
|
256 |
+
"filepath": os.path.join(data_dir, "data"),
|
257 |
+
"metafile": os.path.join(data_dir, _METAFILE),
|
258 |
+
},
|
259 |
+
),
|
260 |
+
# datasets.SplitGenerator(
|
261 |
+
# name=datasets.Split.TEST,
|
262 |
+
# gen_kwargs={"filepath": os.path.join(data_dir, "data"), "metafile": os.path.join(data_dir, _METAFILE)},
|
263 |
+
# ),
|
264 |
+
# datasets.SplitGenerator(
|
265 |
+
# name=datasets.Split.VALIDATION,
|
266 |
+
# gen_kwargs={"filepath": os.path.join(data_dir, "data"), "metafile": os.path.join(data_dir, _METAFILE)},
|
267 |
+
# ),
|
268 |
+
]
|
269 |
+
|
270 |
+
def _generate_examples(self, filepath, metafile, stratify_column: str = None, fold: List = None):
|
271 |
"""
|
272 |
This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
|
273 |
:param filepath: The path where the data is located.
|
274 |
:param metafile: The metafile describing the chall data
|
275 |
+
:param stratify_column: The meta column to stratify by.
|
276 |
+
:param fold: A list of values do define splits. Only works in combination with `stratify_column`
|
277 |
:return:
|
278 |
"""
|
279 |
|
|
|
282 |
with open(metafile, 'r') as file:
|
283 |
metadata = json.load(file)
|
284 |
for row in metadata["data"]:
|
285 |
+
|
286 |
+
# define splits if set
|
287 |
+
if stratify_column and str(row[stratify_column]) not in fold:
|
288 |
+
continue
|
289 |
+
|
290 |
# load transcript
|
291 |
transcript_file = os.path.join(filepath, row["transcript_file"])
|
292 |
with open(transcript_file, 'r') as transcript:
|
|
|
422 |
|
423 |
split_segments = []
|
424 |
for segment in segments:
|
425 |
+
if any(w["end"] - w["start"] >= self.config.max_pause_length and w["text"].strip() == "(...)" for w in segment["words"]):
|
426 |
start_i = 0
|
427 |
for i, word in enumerate(segment["words"]):
|
428 |
w_duration = word["end"] - word["start"]
|
429 |
if w_duration >= self.config.max_pause_length and word["text"].strip() == "(...)":
|
430 |
if len(segment["words"][start_i:i]) > 0:
|
431 |
split_segments.append({"speaker": segment["speaker"], "words": segment["words"][start_i:i]})
|
432 |
+
start_i = i + 1
|
433 |
|
434 |
if len(segment["words"][start_i:]) > 0:
|
435 |
split_segments.append({"speaker": segment["speaker"], "words": segment["words"][start_i:]})
|