29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
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
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598 | class IITBHGCProcessor(DatasetProcessor):
"""Processor for IITBHGC dataset"""
# Text fixes mapping as class constant for better performance
TEXT_FIXES = {
3: [('with Andy', 'with__NBWS__Andy')],
5: [('watch Jose', 'watch__NBWS__Jose')],
9: [('$1,750', '$__NBWS__1,750.00')],
25: [('£3', '£__NBWS__3.00')],
26: [('for Virgil', 'for__NBWS__Virgil')],
33: [('at FC', 'at__NBWS__FC')],
53: [('$5,000', '$__NBWS__5,000.00')],
74: [('$20,000', '$__NBWS__20,000.00')],
82: [('£750,000', '£__NBWS__7,50,000.00')],
99: [('$5.3', '$__NBWS__5.30')],
102: [('$10', '$__NBWS__10.00'), ('$9', '$__NBWS__9.00')],
130: [('$50,000', '$__NBWS__50,000.00')],
257: [('$10', '$__NBWS__10.00')],
280: [('Claim: Mile Jedinak twisted', 'Claim: Mile Jedinak__NBWS__twisted')],
288: [('$2', '$__NBWS__2.00')],
298: [('$200', '$__NBWS__200.00')],
325: [('$9', '$__NBWS__9.00')],
357: [('$1.8', '$__NBWS__1.80'), ('$2.6', '$__NBWS__2.60')],
365: [('$25,000', '$__NBWS__25,000.00')],
373: [
('$260', '$__NBWS__260.00'),
('$1.7', '$__NBWS__1.70'),
('$1.37', '$__NBWS__1.37'),
],
403: [('hour-long', 'hour-long hour-long')],
404: [('$2.9', '$__NBWS__2.90')],
425: [
('- many', '-__NBWS__many'),
('against Leicester City', 'against__NBWS__Leicester__NBWS__City'),
("mock Louis van Gaal's", "mock Louis van__NBWS__Gaal's"),
],
441: [('$10,000', '$__NBWS__10,000.00')],
460: [('$1.6', '$__NBWS__1.60')],
468: [('$105', '$__NBWS__105.00')],
483: [('(£4,943)', '(£__NBWS__4,943.00)')],
485: [('long-running', 'long-running long-running')],
}
@staticmethod
def fix_texts(text: str, text_name: np.int64) -> str:
"""Apply text-specific fixes based on text_name."""
if text_name in IITBHGCProcessor.TEXT_FIXES:
for old, new in IITBHGCProcessor.TEXT_FIXES[text_name]:
text = text.replace(old, new)
return text
@staticmethod
def init_word_dict(text_strs: list[str], text_aois: list[int]) -> dict:
"""Initialize word dictionary with default values."""
return {
int(word_index): {
'IA_LABEL': word.replace('__NBWS__', '\xa0'),
'IA_ID': word_index,
**{
key: 0
for key in [
'FFD',
'SFD',
'FD',
'FPRT',
'FRT',
'TFT',
'RRT',
'RPD_inc',
'RPD_exc',
'RBRT',
'Fix',
'FPF',
'RR',
'FPReg',
'TRC_out',
'TRC_in',
'SL_in',
'SL_out',
'TFC',
]
},
}
for word_index, word in zip(text_aois, text_strs)
}
def compute_word_level_reading_measures(
self,
fix_df: pd.DataFrame,
stim_df: pd.DataFrame,
) -> pd.DataFrame:
"""Compute word-level reading measures from fixation data."""
def process_participant(
text_name: np.int64, participant_id: str
) -> pd.DataFrame:
try:
# Get relevant data slices
aoi_df = stim_df[stim_df[Fields.UNIQUE_PARAGRAPH_ID] == text_name]
tmp_df = fix_df[fix_df[Fields.UNIQUE_PARAGRAPH_ID] == text_name]
fixations_df = tmp_df[
tmp_df[Fields.SUBJECT_ID] == participant_id
].copy()
if fixations_df.empty:
return pd.DataFrame()
assert len(fixations_df.label.unique()) == 1
label = fixations_df['label'].iloc[0]
# Add dummy row
fixations_df = pd.concat(
[
fixations_df,
pd.DataFrame(
[[0] * len(fixations_df.columns)],
columns=fixations_df.columns,
),
],
ignore_index=True,
)
# Process text
text = self.fix_texts(aoi_df.paragraph.iloc[0], text_name)
text_strs = text.split()
text_aois = list(range(len(text_strs)))
# Initialize word dictionary
word_dict = self.init_word_dict(text_strs, text_aois)
# Process fixations
right_most_word = cur_fix_word_idx = next_fix_word_idx = (
next_fix_dur
) = -1
for _, fixation in fixations_df.iterrows():
# Skip invalid fixations
try:
aoi = int(fixation['CURRENT_FIX_X'])
except (ValueError, TypeError):
continue
if (
fixation['IA_LABEL'] == '.'
or fixation['CURRENT_FIX_DURATION'] == 0
):
continue
# Update fixation indices
last_fix_word_idx = cur_fix_word_idx
cur_fix_word_idx = next_fix_word_idx
cur_fix_dur = next_fix_dur
next_fix_word_idx = aoi
next_fix_dur = fixation['CURRENT_FIX_DURATION']
# Validate word match
if aoi in word_dict:
self._validate_word_match(
word_dict[aoi]['IA_LABEL'],
fixation['IA_LABEL'],
text_name,
participant_id,
aoi,
)
else:
continue
if next_fix_dur == 0:
next_fix_word_idx = cur_fix_word_idx
if cur_fix_word_idx == -1:
continue
right_most_word = max(right_most_word, cur_fix_word_idx)
# Update word statistics
cur_word = word_dict[cur_fix_word_idx]
cur_word['TFT'] += int(cur_fix_dur)
cur_word['TFC'] += 1
if cur_word['FD'] == 0:
cur_word['FD'] = int(cur_fix_dur)
if right_most_word == cur_fix_word_idx:
if cur_word['TRC_out'] == 0:
cur_word['FPRT'] += int(cur_fix_dur)
if last_fix_word_idx < cur_fix_word_idx:
cur_word['FFD'] += int(cur_fix_dur)
else:
word_dict[right_most_word]['RPD_exc'] += int(cur_fix_dur)
if cur_fix_word_idx < last_fix_word_idx:
cur_word['TRC_in'] += 1
if cur_fix_word_idx > next_fix_word_idx:
cur_word['TRC_out'] += 1
if cur_fix_word_idx == right_most_word:
cur_word['RBRT'] += int(cur_fix_dur)
if cur_word['FRT'] == 0 and (
cur_fix_word_idx != next_fix_word_idx or next_fix_dur == 0
):
cur_word['FRT'] = cur_word['TFT']
if cur_word['SL_in'] == 0:
cur_word['SL_in'] = cur_fix_word_idx - last_fix_word_idx
if cur_word['SL_out'] == 0:
cur_word['SL_out'] = next_fix_word_idx - cur_fix_word_idx
# Finalize word measures
rows = []
for word_index, word_rm in word_dict.items():
if word_rm['FFD'] == word_rm['FPRT']:
word_rm['SFD'] = word_rm['FFD']
word_rm['RRT'] = word_rm['TFT'] - word_rm['FPRT']
word_rm['FPF'] = int(word_rm['FFD'] > 0)
word_rm['RR'] = int(word_rm['RRT'] > 0)
word_rm['FPReg'] = int(word_rm['RPD_exc'] > 0)
word_rm['Fix'] = int(word_rm['TFT'] > 0)
word_rm['RPD_inc'] = word_rm['RPD_exc'] + word_rm['RBRT']
word_rm[Fields.SUBJECT_ID] = participant_id
word_rm[Fields.UNIQUE_PARAGRAPH_ID] = text_name
word_rm['paragraph'] = text
word_rm['word_index'] = word_index
word_rm['label'] = label
rows.append(word_rm)
return pd.DataFrame(rows)
except Exception as e:
logger.exception(
f'Error processing {text_name} - {participant_id}: {e}'
)
return pd.DataFrame()
# Process all participants in parallel-ready structure
rm_df_parts = [
process_participant(text_name, participant_id)
for text_name in fix_df[Fields.UNIQUE_PARAGRAPH_ID].unique()
for participant_id in fix_df[
fix_df[Fields.UNIQUE_PARAGRAPH_ID] == text_name
][Fields.SUBJECT_ID].unique()
]
return pd.concat(rm_df_parts, ignore_index=True)
def _validate_word_match(
self,
expected: str,
actual: str,
text_name: np.int64,
participant_id: str,
aoi: int,
) -> None:
"""Validate word match between expected and actual labels."""
if expected == actual:
return
# Check for known acceptable mismatches
if (
actual in expected
or expected.strip('(').strip(')')
in actual # problems in their aois vs. paragraph
or expected.lower() == actual.lower() # case differences
or text_name
in [122, 305] # known issues in these texts [wrong kommata in numbers]
):
logger.info(f'Acceptable mismatch: "{expected}" vs "{actual}"')
return
if '\xa0' in actual:
logger.warning(
f'Non-breaking space in fixation: text={text_name}, participant={participant_id}, aoi={aoi}'
)
logger.warning(
f'Mismatch in text {text_name} for participant {participant_id} at AOI {aoi}: '
f'expected "{expected}", got "{actual}"'
)
def get_column_map(self, data_type: DataType) -> dict:
"""Get column mapping for IITBHGC dataset."""
base_map = {
'trial_id': Fields.UNIQUE_PARAGRAPH_ID,
'participant_id': Fields.SUBJECT_ID,
'fixation_word_ids': 'CURRENT_FIX_X',
'fixation_durations': 'CURRENT_FIX_DURATION',
'text': 'paragraph',
'fixation_seqs': 'CURRENT_FIX_INDEX',
'fixation_word_texts': 'IA_LABEL',
}
return base_map if data_type in [DataType.IA, DataType.FIXATIONS] else {}
def get_columns_to_keep(self) -> list:
"""Get list of columns to keep after filtering."""
return []
def dataset_specific_processing(
self, data_dict: dict[str, pd.DataFrame]
) -> dict[str, pd.DataFrame]:
"""IITBHGC-specific processing steps."""
# Process IA and FIXATIONS data
for data_type in [DataType.IA, DataType.FIXATIONS]:
if data_type not in data_dict or data_dict[data_type] is None:
continue
df = data_dict[data_type]
# Add unique trial ID
df[Fields.UNIQUE_TRIAL_ID] = (
df[Fields.SUBJECT_ID].astype(str)
+ '_'
+ df[Fields.UNIQUE_PARAGRAPH_ID].astype(str)
)
# Parse list columns
list_columns = [
'CURRENT_FIX_X',
'CURRENT_FIX_DURATION',
'CURRENT_FIX_INDEX',
'IA_LABEL',
]
df[list_columns] = df[list_columns].map(literal_eval)
df = df.explode(list_columns).reset_index(drop=True)
# Clean text
df['paragraph'] = (
df['paragraph']
.str.replace('↵', ' ', regex=False)
.str.split()
.str.join(' ')
)
df['label'] = (df['true_labels'] == df['annotator_labels']).astype(int)
df['CURRENT_FIX_X'] = df['CURRENT_FIX_X'].apply(
lambda x: int(x) - 1 if x != '.' else -1
)
if data_type == DataType.FIXATIONS:
df['CURRENT_FIX_Y'] = 0
df['CURRENT_FIX_INDEX'] = df['CURRENT_FIX_INDEX'].astype(int)
data_dict[data_type] = df
# Compute reading measures
rm_df = self.compute_word_level_reading_measures(
data_dict['fixations'], data_dict['ia']
)
data_dict['ia'] = rm_df
# Add IA features to fixation data
logger.info('Adding IA report features to fixation data...')
data_dict[DataType.FIXATIONS], data_dict[DataType.IA] = (
self.add_ia_report_features_to_fixation_data(
data_dict[DataType.IA],
data_dict[DataType.FIXATIONS],
)
)
# Add missing features
for data_type in [DataType.IA, DataType.FIXATIONS]:
if data_type == DataType.IA:
data_dict['ia']['NEXT_FIX_INTEREST_AREA_INDEX'] = data_dict['ia'][
'word_index'
].shift(-1)
data_dict['ia']['PREVIOUS_FIX_INTEREST_AREA_INDEX'] = data_dict['ia'][
'word_index'
].shift(1)
else:
data_dict['fixations']['NEXT_FIX_INTEREST_AREA_INDEX'] = 0
data_dict['fixations']['NEXT_SAC_PEAK_VELOCITY'] = 0
data_dict[data_type] = add_missing_features(
et_data=data_dict[data_type],
trial_groupby_columns=self.data_args.groupby_columns,
mode=data_type,
)
# Compute trial-level features
trial_level_features = compute_trial_level_features(
raw_fixation_data=data_dict[DataType.FIXATIONS],
raw_ia_data=data_dict[DataType.IA],
trial_groupby_columns=self.data_args.groupby_columns,
processed_data_path=self.data_args.processed_data_path,
)
data_dict[DataType.TRIAL_LEVEL] = trial_level_features
return data_dict
def add_ia_report_features_to_fixation_data(
self, ia_df: pd.DataFrame, fix_df: pd.DataFrame
) -> tuple[pd.DataFrame, pd.DataFrame]:
"""Merge per-IA features into fixation-level data."""
# Remove duplicates from groupby columns
self.data_args.groupby_columns = list(
dict.fromkeys(self.data_args.groupby_columns)
)
# Rename IA ID column
ia_df = ia_df.rename(
columns={
Fields.IA_DATA_IA_ID_COL_NAME: Fields.FIXATION_REPORT_IA_ID_COL_NAME
}
)
# Add computed columns
ia_df['unique_trial_id'] = (
ia_df['participant_id'].astype(str)
+ '_'
+ ia_df['unique_paragraph_id'].astype(str)
)
ia_df['word_length'] = ia_df['IA_LABEL'].str.len()
ia_df['TRIAL_IA_COUNT'] = ia_df.groupby('unique_trial_id')[
'unique_trial_id'
].transform('count')
surp_extractor = get_surp_extractor(
extractor_type=SurpExtractorType.CAT_CTX_LEFT, model_name='gpt2'
)
nlp = spacy.load('en_core_web_sm')
# Process metrics
def process_group(group):
sentence = group.iloc[0].paragraph
metrics = get_metrics(
target_text=sentence,
surp_extractor=surp_extractor,
parsing_model=nlp,
parsing_mode='re-tokenize',
add_parsing_features=True,
language='en',
)
metrics['unique_paragraph_id'] = group['unique_paragraph_id'].iloc[0]
metrics[Fields.FIXATION_REPORT_IA_ID_COL_NAME] = metrics['Token_idx']
return metrics
metrics_list = [
process_group(group)
for _, group in tqdm(
ia_df.groupby(Fields.UNIQUE_PARAGRAPH_ID), desc='Processing metrics'
)
]
metrics_df = pd.concat(metrics_list, ignore_index=True)
# Merge metrics
ia_df[Fields.UNIQUE_TRIAL_ID] = (
ia_df[Fields.SUBJECT_ID].astype(str)
+ '_'
+ ia_df[Fields.UNIQUE_PARAGRAPH_ID].astype(str)
)
merge_keys = {'unique_paragraph_id', Fields.FIXATION_REPORT_IA_ID_COL_NAME}
drop_keys = (set(metrics_df.columns) & set(ia_df.columns)) - merge_keys
ia_df['CURRENT_FIX_INTEREST_AREA_INDEX'] = ia_df['word_index']
ia_df = ia_df.merge(
metrics_df.drop(columns=list(drop_keys)), on=list(merge_keys), how='left'
)
# Rename columns
column_renames = {
'POS': 'universal_pos',
'Length': 'word_length_no_punctuation',
'Wordfreq_Frequency': 'wordfreq_frequency',
'subtlex_Frequency': 'subtlex_frequency',
'Reduced_POS': 'ptb_pos',
'Head_word_idx': 'head_word_index',
'Dependency_Relation': 'dependency_relation',
'Entity': 'entity_type',
'gpt2_Surprisal': 'gpt2_surprisal',
'gpt2': 'gpt2_surprisal',
'Head_Direction': 'head_direction',
'Is_Content_Word': 'is_content_word',
'n_Lefts': 'left_dependents_count',
'n_Rights': 'right_dependents_count',
'Distance2Head': 'distance_to_head',
}
ia_df = ia_df.rename(columns=column_renames)
# Add default columns efficiently
zero_columns = [
'start_of_line',
'end_of_line',
'IA_LAST_FIXATION_DURATION',
'IA_LAST_RUN_DWELL_TIME',
'IA_SELECTIVE_REGRESSION_PATH_DURATION',
'IA_FIRST_FIXATION_VISITED_IA_COUNT',
'IA_LEFT',
'IA_RIGHT',
'IA_TOP',
'IA_BOTTOM',
'IA_REGRESSION_PATH_DURATION',
'IA_REGRESSION_OUT_COUNT',
'IA_REGRESSION_IN_COUNT',
'IA_FIRST_FIX_PROGRESSIVE',
'normalized_ID',
'IA_FIRST_RUN_FIXATION_COUNT',
'IA_LAST_RUN_FIXATION_COUNT',
]
ia_df[zero_columns] = 0
# Computed columns
ia_df['IA_FIRST_RUN_DWELL_TIME'] = ia_df['FPRT']
ia_df['IA_FIRST_RUN_FIXATION_DURATION'] = ia_df['FPRT']
ia_df['IA_DWELL_TIME'] = ia_df['FD']
ia_df['IA_DWELL_TIME_%'] = ia_df.groupby('unique_trial_id')[
'IA_DWELL_TIME'
].transform(lambda x: x / x.sum() if x.sum() > 0 else 0)
ia_df['PARAGRAPH_RT'] = ia_df.groupby(Fields.UNIQUE_PARAGRAPH_ID)[
'IA_DWELL_TIME'
].transform('sum')
ia_df['IA_SKIP'] = (ia_df['Fix'] > 0).astype(int)
ia_df['total_skip'] = (ia_df['Fix'] > 0).astype(int)
ia_df['IA_FIXATION_COUNT'] = ia_df['TFC']
ia_df['IA_FIXATION_%'] = ia_df.groupby('unique_trial_id')[
'IA_FIXATION_COUNT'
].transform(lambda x: x / np.sum(x))
ia_df['IA_FIRST_FIXATION_DURATION'] = ia_df['FFD']
ia_df['IA_SINGLE_FIXATION_DURATION'] = ia_df['SFD']
ia_df['IA_RUN_COUNT'] = ia_df['TFC']
ia_df['IA_REGRESSION_OUT_FULL_COUNT'] = ia_df['TRC_out']
# Fixation defaults
fix_df['CURRENT_FIX_PUPIL'] = 0
fix_df['CURRENT_FIX_NEAREST_INTEREST_AREA_DISTANCE'] = (
fix_df['IA_LABEL'] == '.'
).astype(int)
fix_df[['NEXT_SAC_DURATION', 'NEXT_SAC_AVG_VELOCITY', 'NEXT_SAC_AMPLITUDE']] = 0
fix_df['CURRENT_FIX_INTEREST_AREA_INDEX'] = fix_df['CURRENT_FIX_X'].fillna(-1)
# IA defaults for spatial/angular features
spatial_columns = [
'NEXT_SAC_START_X',
'NEXT_SAC_END_X',
'NEXT_SAC_END_Y',
'NEXT_SAC_START_Y',
'PREVIOUS_FIX_DISTANCE',
'NEXT_SAC_ANGLE',
'NEXT_FIX_ANGLE',
'NEXT_FIX_DISTANCE',
'PREVIOUS_FIX_ANGLE',
]
ia_df[spatial_columns] = 0
# Merge fixations with IA features
merge_keys = set(
self.data_args.groupby_columns + [Fields.FIXATION_REPORT_IA_ID_COL_NAME]
)
dup_cols = (set(fix_df.columns) & set(ia_df.columns)) - merge_keys
_ia_df = ia_df.drop(columns=list(dup_cols))
if 'normalized_part_ID' in fix_df.columns:
fix_df = fix_df.drop(columns='normalized_part_ID')
enriched_fix_df = fix_df.merge(
_ia_df,
on=list(merge_keys),
how='left',
validate='many_to_one',
)
# Add word count
num_of_words_in_trials = ia_df.groupby(self.data_args.groupby_columns).size()
num_of_words_in_trials.name = 'num_of_words_in_trial'
enriched_fix_df = enriched_fix_df.merge(
num_of_words_in_trials,
on=self.data_args.groupby_columns,
how='left',
)
return enriched_fix_df, ia_df
|