Datasets:
Change re config to match format described in notion notebook and add option to process on document level
Browse files
mobie.py
CHANGED
@@ -53,6 +53,7 @@ _URLs = {
|
|
53 |
|
54 |
|
55 |
def simplify_dict(d, remove_attribute=True):
|
|
|
56 |
if isinstance(d, dict):
|
57 |
new_dict = {}
|
58 |
for k, v in d.items():
|
@@ -123,7 +124,7 @@ class Mobie(datasets.GeneratorBasedBuilder):
|
|
123 |
"time",
|
124 |
"trigger",
|
125 |
]
|
126 |
-
|
127 |
{
|
128 |
"id": datasets.Value("string"),
|
129 |
"text": datasets.Value("string"),
|
@@ -158,44 +159,31 @@ class Mobie(datasets.GeneratorBasedBuilder):
|
|
158 |
{
|
159 |
"id": datasets.Value("string"),
|
160 |
"text": datasets.Value("string"),
|
161 |
-
"
|
162 |
}
|
163 |
)
|
164 |
elif self.config.name == "re":
|
165 |
features = datasets.Features(
|
166 |
{
|
167 |
"id": datasets.Value("string"),
|
168 |
-
"
|
169 |
-
"
|
170 |
-
"
|
171 |
-
|
172 |
-
"
|
173 |
-
"
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
names=[
|
187 |
-
"no_arg", "location", "delay", "direction", "start_loc", "end_loc",
|
188 |
-
"start_date", "end_date", "cause", "jam_length", "route"
|
189 |
-
]
|
190 |
-
),
|
191 |
-
"event_type": datasets.features.ClassLabel(
|
192 |
-
names=[
|
193 |
-
"O", "Accident", "CanceledRoute", "CanceledStop", "Delay", "Obstruction",
|
194 |
-
"RailReplacementService", "TrafficJam"
|
195 |
-
]
|
196 |
-
)
|
197 |
-
}
|
198 |
-
]
|
199 |
}
|
200 |
)
|
201 |
elif self.config.name == "ee":
|
@@ -204,7 +192,7 @@ class Mobie(datasets.GeneratorBasedBuilder):
|
|
204 |
{
|
205 |
"id": datasets.Value("string"),
|
206 |
"text": datasets.Value("string"),
|
207 |
-
"entity_mentions":
|
208 |
"event_mentions": [
|
209 |
{
|
210 |
"id": datasets.Value("string"),
|
@@ -284,7 +272,7 @@ class Mobie(datasets.GeneratorBasedBuilder):
|
|
284 |
),
|
285 |
]
|
286 |
|
287 |
-
def _generate_examples(self, filepath, split):
|
288 |
"""Yields examples."""
|
289 |
|
290 |
if self.config.name == "ner":
|
@@ -307,11 +295,8 @@ class Mobie(datasets.GeneratorBasedBuilder):
|
|
307 |
|
308 |
for doc in decode_stacked(raw):
|
309 |
text = doc["text"]["string"]
|
310 |
-
|
311 |
-
|
312 |
-
# for m in doc["conceptMentions"]["array"]:
|
313 |
-
# entity_starts.append(m["span"]["start"])
|
314 |
-
for s in doc["sentences"]["array"]:
|
315 |
toks = []
|
316 |
lbls = []
|
317 |
sid = s["id"]
|
@@ -331,16 +316,17 @@ class Mobie(datasets.GeneratorBasedBuilder):
|
|
331 |
doc = json.loads(line)
|
332 |
doc = simplify_dict(doc)
|
333 |
text = doc["text"]
|
334 |
-
|
|
|
335 |
sentence_id = sentence["id"]
|
336 |
-
sentence_start = sentence["span"]["start"]
|
337 |
mobie_cms = sentence["conceptMentions"]
|
338 |
-
|
339 |
for cm in mobie_cms:
|
340 |
cm_start = cm["span"]["start"]
|
341 |
cm_end = cm["span"]["end"]
|
342 |
cm_text = text[cm_start:cm_end]
|
343 |
-
|
344 |
"id": cm["id"],
|
345 |
"text": cm_text,
|
346 |
"start": cm_start - sentence_start,
|
@@ -354,58 +340,72 @@ class Mobie(datasets.GeneratorBasedBuilder):
|
|
354 |
] if "refids" in cm and cm["refids"] else []
|
355 |
})
|
356 |
if self.config.name == "el":
|
|
|
357 |
yield sentence_id, {
|
358 |
"id": sentence_id,
|
359 |
"text": text,
|
360 |
-
"
|
361 |
}
|
362 |
elif self.config.name == "re":
|
363 |
mobie_rms = sentence["relationMentions"]
|
364 |
if not mobie_rms:
|
365 |
continue
|
366 |
-
|
367 |
-
|
368 |
-
|
369 |
-
|
370 |
-
|
371 |
-
|
372 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
373 |
break
|
374 |
-
if
|
375 |
-
|
376 |
-
|
377 |
-
|
378 |
-
trigger_text = text[trigger_start:trigger_end]
|
379 |
for arg in rm["args"]:
|
380 |
-
|
381 |
-
|
382 |
-
|
383 |
-
|
384 |
-
|
385 |
-
|
386 |
-
|
387 |
-
"
|
388 |
-
|
389 |
-
|
390 |
-
|
391 |
-
|
392 |
-
|
393 |
-
|
394 |
-
|
395 |
-
|
396 |
-
|
397 |
-
|
398 |
-
|
399 |
-
|
400 |
-
|
401 |
-
|
402 |
-
|
403 |
-
|
404 |
-
|
405 |
-
|
406 |
-
|
407 |
-
|
408 |
-
|
409 |
elif self.config.name == "ee":
|
410 |
mobie_rms = sentence["relationMentions"]
|
411 |
if not mobie_rms:
|
@@ -452,7 +452,7 @@ class Mobie(datasets.GeneratorBasedBuilder):
|
|
452 |
yield sentence_id, {
|
453 |
"id": sentence_id,
|
454 |
"text": text,
|
455 |
-
"entity_mentions":
|
456 |
"event_mentions": event_mentions
|
457 |
}
|
458 |
else:
|
|
|
53 |
|
54 |
|
55 |
def simplify_dict(d, remove_attribute=True):
|
56 |
+
"""Simplify nested dictionary by removing unnecessary keys"""
|
57 |
if isinstance(d, dict):
|
58 |
new_dict = {}
|
59 |
for k, v in d.items():
|
|
|
124 |
"time",
|
125 |
"trigger",
|
126 |
]
|
127 |
+
entity_mentions = [
|
128 |
{
|
129 |
"id": datasets.Value("string"),
|
130 |
"text": datasets.Value("string"),
|
|
|
159 |
{
|
160 |
"id": datasets.Value("string"),
|
161 |
"text": datasets.Value("string"),
|
162 |
+
"entity_mentions": entity_mentions
|
163 |
}
|
164 |
)
|
165 |
elif self.config.name == "re":
|
166 |
features = datasets.Features(
|
167 |
{
|
168 |
"id": datasets.Value("string"),
|
169 |
+
"tokens": datasets.Sequence(datasets.Value("string")),
|
170 |
+
"entities": datasets.Sequence([datasets.Value("int32")]),
|
171 |
+
"entity_roles": datasets.Sequence(datasets.features.ClassLabel(
|
172 |
+
names=[
|
173 |
+
"no_arg", "trigger", "location", "delay", "direction", "start_loc", "end_loc",
|
174 |
+
"start_date", "end_date", "cause", "jam_length", "route"
|
175 |
+
]
|
176 |
+
)),
|
177 |
+
"entity_types": datasets.Sequence(datasets.features.ClassLabel(
|
178 |
+
names=labels
|
179 |
+
)),
|
180 |
+
"event_type": datasets.features.ClassLabel(
|
181 |
+
names=[
|
182 |
+
"O", "Accident", "CanceledRoute", "CanceledStop", "Delay", "Obstruction",
|
183 |
+
"RailReplacementService", "TrafficJam"
|
184 |
+
]
|
185 |
+
),
|
186 |
+
"entity_ids": datasets.Sequence(datasets.Value("string"))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
187 |
}
|
188 |
)
|
189 |
elif self.config.name == "ee":
|
|
|
192 |
{
|
193 |
"id": datasets.Value("string"),
|
194 |
"text": datasets.Value("string"),
|
195 |
+
"entity_mentions": entity_mentions,
|
196 |
"event_mentions": [
|
197 |
{
|
198 |
"id": datasets.Value("string"),
|
|
|
272 |
),
|
273 |
]
|
274 |
|
275 |
+
def _generate_examples(self, filepath, split, sentence_level=False):
|
276 |
"""Yields examples."""
|
277 |
|
278 |
if self.config.name == "ner":
|
|
|
295 |
|
296 |
for doc in decode_stacked(raw):
|
297 |
text = doc["text"]["string"]
|
298 |
+
iterable = doc["sentences"]["array"] if sentence_level else [doc]
|
299 |
+
for s in iterable:
|
|
|
|
|
|
|
300 |
toks = []
|
301 |
lbls = []
|
302 |
sid = s["id"]
|
|
|
316 |
doc = json.loads(line)
|
317 |
doc = simplify_dict(doc)
|
318 |
text = doc["text"]
|
319 |
+
iterable = doc["sentences"] if sentence_level else [doc]
|
320 |
+
for sentence in iterable:
|
321 |
sentence_id = sentence["id"]
|
322 |
+
sentence_start = sentence["span"]["start"] if "span" in sentence else 0
|
323 |
mobie_cms = sentence["conceptMentions"]
|
324 |
+
entity_mentions = []
|
325 |
for cm in mobie_cms:
|
326 |
cm_start = cm["span"]["start"]
|
327 |
cm_end = cm["span"]["end"]
|
328 |
cm_text = text[cm_start:cm_end]
|
329 |
+
entity_mentions.append({
|
330 |
"id": cm["id"],
|
331 |
"text": cm_text,
|
332 |
"start": cm_start - sentence_start,
|
|
|
340 |
] if "refids" in cm and cm["refids"] else []
|
341 |
})
|
342 |
if self.config.name == "el":
|
343 |
+
# TODO use osm_id as entity id?
|
344 |
yield sentence_id, {
|
345 |
"id": sentence_id,
|
346 |
"text": text,
|
347 |
+
"entity_mentions": entity_mentions
|
348 |
}
|
349 |
elif self.config.name == "re":
|
350 |
mobie_rms = sentence["relationMentions"]
|
351 |
if not mobie_rms:
|
352 |
continue
|
353 |
+
tokens = [text[token["span"]["start"]:token["span"]["end"]] for token in sentence["tokens"]]
|
354 |
+
entities = []
|
355 |
+
entity_types = []
|
356 |
+
entity_ids = []
|
357 |
+
for cm in mobie_cms:
|
358 |
+
# Find token offsets for entity mentions
|
359 |
+
start = -1
|
360 |
+
end = -1
|
361 |
+
for idx, token in enumerate(sentence["tokens"]):
|
362 |
+
if token["span"]["start"] == cm["span"]["start"]:
|
363 |
+
start = idx
|
364 |
+
if token["span"]["end"] == cm["span"]["end"]:
|
365 |
+
end = idx
|
366 |
+
assert start != -1 and end != -1, f"Could not find token offsets for {cm['id']}"
|
367 |
+
entities.append([start, end])
|
368 |
+
entity_types.append(cm["type"])
|
369 |
+
found_osm_id = False
|
370 |
+
for refid in cm["refids"]:
|
371 |
+
if refid["key"] == "osm_id":
|
372 |
+
entity_ids.append(refid["value"])
|
373 |
+
found_osm_id = True
|
374 |
break
|
375 |
+
if not found_osm_id:
|
376 |
+
entity_ids.append("NIL")
|
377 |
+
for rm in mobie_rms:
|
378 |
+
entity_roles = ["no_arg"] * len(entities)
|
|
|
379 |
for arg in rm["args"]:
|
380 |
+
entity_role = arg["role"]
|
381 |
+
# Matching via ids does not work, need to match via position
|
382 |
+
# Find token offsets for entity mentions
|
383 |
+
start = -1
|
384 |
+
end = -1
|
385 |
+
cm = arg["conceptMention"]
|
386 |
+
for idx, token in enumerate(sentence["tokens"]):
|
387 |
+
if token["span"]["start"] == cm["span"]["start"]:
|
388 |
+
start = idx
|
389 |
+
if token["span"]["end"] == cm["span"]["end"]:
|
390 |
+
end = idx
|
391 |
+
assert start != -1 and end != -1, f"Could not find token offsets for {cm['id']}"
|
392 |
+
entity_idx = -1
|
393 |
+
for idx, entity in enumerate(entities):
|
394 |
+
if entity == [start, end]:
|
395 |
+
entity_idx = idx
|
396 |
+
break
|
397 |
+
assert entity_idx != -1, f"Could not find entity for {cm['id']}"
|
398 |
+
entity_roles[entity_idx] = entity_role
|
399 |
+
yield f"{sentence_id}_{example_idx}", {
|
400 |
+
"id": f"{sentence_id}_{example_idx}",
|
401 |
+
"tokens": tokens,
|
402 |
+
"entities": entities,
|
403 |
+
"entity_roles": entity_roles,
|
404 |
+
"entity_types": entity_types,
|
405 |
+
"event_type": rm["name"],
|
406 |
+
"entity_ids": entity_ids
|
407 |
+
}
|
408 |
+
example_idx += 1
|
409 |
elif self.config.name == "ee":
|
410 |
mobie_rms = sentence["relationMentions"]
|
411 |
if not mobie_rms:
|
|
|
452 |
yield sentence_id, {
|
453 |
"id": sentence_id,
|
454 |
"text": text,
|
455 |
+
"entity_mentions": entity_mentions,
|
456 |
"event_mentions": event_mentions
|
457 |
}
|
458 |
else:
|