27
28
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 | class PoTeCProcessor(DatasetProcessor):
"""Processor for PoTeC dataset with optimized performance"""
IA_COLUMN_MAP = {
'word': 'IA_LABEL',
'FFD': 'IA_FIRST_FIXATION_DURATION',
'text_surprisal_gpt2-base': 'gpt2_surprisal',
'RPD_in': 'IA_REGRESSION_IN_COUNT',
'document_frequency_normalized': 'wordfreq_frequency',
'text_id': Fields.UNIQUE_PARAGRAPH_ID,
'PoS_tag': 'universal_pos',
'FD': 'IA_DWELL_TIME',
'TFC': 'IA_FIXATION_COUNT',
'FPRT': 'IA_FIRST_RUN_DWELL_TIME',
'word_index_in_text': Fields.IA_DATA_IA_ID_COL_NAME,
'reader_id': 'participant_id',
}
FIXATION_COLUMN_MAP = {
'fixation_index': 'CURRENT_FIX_INDEX',
'fixation_duration': 'CURRENT_FIX_DURATION',
'fixation_position_x': 'CURRENT_FIX_X',
'fixation_position_y': 'CURRENT_FIX_Y',
'aoi': 'IA_ID',
'next_saccade_duration': 'NEXT_SAC_DURATION',
'reader_id': Fields.SUBJECT_ID,
'item_id': Fields.UNIQUE_PARAGRAPH_ID,
}
LINGUISTIC_FEATURE_RENAMES = {
'POS': 'universal_pos',
'Length': 'word_length_no_punctuation',
'subtlex_Frequency': 'subtlex_frequency',
'Reduced_POS': 'ptb_pos',
'Head_word_idx': 'head_word_index',
'Dependency_Relation': 'dependency_relation',
'Entity': 'entity_type',
'Head_Direction': 'head_direction',
'Is_Content_Word': 'is_content_word',
'n_Lefts': 'left_dependents_count',
'n_Rights': 'right_dependents_count',
'Distance2Head': 'distance_to_head',
}
def get_column_map(self, data_type: DataType) -> dict[str, str]:
return (
self.IA_COLUMN_MAP if data_type == DataType.IA else self.FIXATION_COLUMN_MAP
)
def get_columns_to_keep(self) -> list[str]:
return []
def _build_stim_dict(self, stim_df: pd.DataFrame) -> dict[str, pd.DataFrame]:
return {name: data for name, data in stim_df.groupby('text_name')}
def _process_single_page(
self,
page_name: str,
page_fixes: pd.DataFrame,
stim_dict: dict[str, pd.DataFrame],
) -> tuple:
if page_name not in stim_dict:
return None, None, None, None, None
page_stim = stim_dict[page_name]
page_indices = page_fixes.index
n_fixes = len(page_indices)
word_values = np.full(n_fixes, '.', dtype=object)
aoi_values = np.full(n_fixes, np.nan)
start_line = np.full(n_fixes, False)
end_line = np.full(n_fixes, False)
fix_coords = page_fixes.loc[
page_indices, ['CURRENT_FIX_X', 'CURRENT_FIX_Y']
].values
for i, (fix_x, fix_y) in enumerate(fix_coords):
x_match = (page_stim['start_x'] <= fix_x) & (fix_x < page_stim['end_x'])
y_match = (page_stim['start_y'] <= fix_y) & (fix_y < page_stim['end_y'])
aoi_match = page_stim[x_match & y_match]
if not aoi_match.empty:
first_match = aoi_match.iloc[0]
word_values[i] = first_match['character']
aoi_values[i] = int(first_match['aoi'])
return page_indices, word_values, aoi_values, start_line, end_line
def map_to_aois(self, fix_df: pd.DataFrame, stim_df: pd.DataFrame) -> pd.DataFrame:
fix_df['word'] = '.'
fix_df['CURRENT_FIX_INTEREST_AREA_INDEX'] = np.nan
fix_df['start_of_line'] = False
fix_df['end_of_line'] = False
stim_dict = self._build_stim_dict(stim_df)
page_groups = list(fix_df.groupby(Fields.UNIQUE_PARAGRAPH_ID))
results = Parallel(n_jobs=-1, backend='threading')(
delayed(self._process_single_page)(page_name, page_fixes, stim_dict)
for page_name, page_fixes in tqdm(
page_groups, desc='Mapping fixations to AOIs'
)
)
for page_indices, word_values, aoi_values, start_line, end_line in results:
if page_indices is not None:
fix_df.loc[page_indices, 'word'] = word_values
fix_df.loc[page_indices, 'CURRENT_FIX_INTEREST_AREA_INDEX'] = aoi_values
fix_df.loc[page_indices, 'start_of_line'] = start_line
fix_df.loc[page_indices, 'end_of_line'] = end_line
return fix_df
def _extract_linguistic_features_for_group(
self, group: pd.DataFrame, nlp, surp_extractor
) -> pd.DataFrame:
"""Extract linguistic features for a single group"""
words = group['IA_LABEL'].fillna('Null').tolist()
sentence = ' '.join(words)
metrics = get_metrics(
target_text=sentence,
parsing_model=nlp,
surp_extractor=surp_extractor,
parsing_mode='re-tokenize',
add_parsing_features=True,
language='de',
)
metrics['unique_paragraph_id'] = group['unique_paragraph_id'].iloc[0]
metrics['unique_trial_id'] = group['unique_trial_id'].iloc[0]
metrics[Fields.FIXATION_REPORT_IA_ID_COL_NAME] = metrics['Token_idx']
return metrics
def _initialize_missing_columns(
self, ia_df: pd.DataFrame, fix_df: pd.DataFrame
) -> tuple[pd.DataFrame, pd.DataFrame]:
ia_zero_cols = [
'IA_FIRST_RUN_FIXATION_COUNT',
'IA_FIRST_FIXATION_VISITED_IA_COUNT',
'IA_REGRESSION_IN_COUNT',
'TRIAL_IA_COUNT',
'IA_SELECTIVE_REGRESSION_PATH_DURATION',
'IA_LAST_RUN_FIXATION_COUNT',
'IA_TOP',
'IA_LAST_RUN_DWELL_TIME',
'IA_REGRESSION_OUT_FULL_COUNT',
'start_of_line',
'end_of_line',
'IA_LEFT',
'IA_LAST_FIXATION_DURATION',
'IA_REGRESSION_PATH_DURATION',
'IA_REGRESSION_OUT_COUNT',
'IA_FIRST_FIX_PROGRESSIVE',
'IA_RUN_COUNT',
'NEXT_SAC_START_Y',
'NEXT_SAC_START_X',
'NEXT_SAC_END_X',
'NEXT_SAC_END_Y',
]
for col in ia_zero_cols:
ia_df[col] = 0
fix_zero_cols = ['NEXT_SAC_AVG_VELOCITY', 'NEXT_SAC_AMPLITUDE']
for col in fix_zero_cols:
fix_df[col] = 0
ia_df['normalized_ID'] = ia_df['CURRENT_FIX_INTEREST_AREA_INDEX']
ia_df['PARAGRAPH_RT'] = ia_df.groupby('unique_trial_id')[
'IA_DWELL_TIME'
].transform('sum')
ia_df['IA_DWELL_TIME_%'] = ia_df.groupby('unique_trial_id')[
'IA_DWELL_TIME'
].transform(lambda x: x / np.sum(x))
ia_df['total_skip'] = (ia_df['IA_DWELL_TIME'] > 0).astype(int)
ia_df['IA_SKIP'] = (ia_df['IA_DWELL_TIME'] > 0).astype(int)
ia_df['IA_FIXATION_%'] = ia_df.groupby('unique_trial_id')[
'IA_FIXATION_COUNT'
].transform(lambda x: x / np.sum(x))
return ia_df, fix_df
def add_ia_report_features_to_fixation_data(
self,
ia_df: pd.DataFrame,
fix_df: pd.DataFrame,
) -> tuple[pd.DataFrame, pd.DataFrame]:
# remove duplicate groupby columns
if len(set(self.data_args.groupby_columns)) != len(
self.data_args.groupby_columns
):
logger.warning('Removing duplicate groupby_columns')
self.data_args.groupby_columns = list(
dict.fromkeys(self.data_args.groupby_columns)
)
# unify IA ID column name
ia_df = ia_df.rename(
columns={
Fields.IA_DATA_IA_ID_COL_NAME: Fields.FIXATION_REPORT_IA_ID_COL_NAME
}
)
# prepare IA dataframe
ia_df['IA_LABEL'] = ia_df['IA_LABEL'].fillna('Null')
paragraphs = ia_df.groupby('unique_trial_id')['IA_LABEL'].apply(
lambda x: ' '.join(x)
)
ia_df['paragraph'] = ia_df['unique_trial_id'].map(paragraphs)
ia_df['CURRENT_FIX_NEAREST_INTEREST_AREA_DISTANCE'] = 0
# extract linguistic features
logger.info(
'Processing linguistic features in batches. This might take a while.'
)
nlp = spacy.load('de_core_news_sm')
surp_extractor = get_surp_extractor(
extractor_type=SurpExtractorType.CAT_CTX_LEFT,
model_name='gpt2',
)
groups = [group for _, group in ia_df.groupby('unique_trial_id')]
metrics_list = []
batch_size = 100
for i in tqdm(range(0, len(groups), batch_size), desc='Processing batches'):
batch = Parallel(n_jobs=-1, backend='threading')(
delayed(self._extract_linguistic_features_for_group)(
g, nlp, surp_extractor
)
for g in groups[i : i + batch_size]
)
metrics_list.extend(batch)
metrics_df = pd.concat(metrics_list, ignore_index=True)
# merge linguistic features
merge_keys = ['unique_trial_id', Fields.FIXATION_REPORT_IA_ID_COL_NAME]
drop_cols = (set(ia_df.columns) & set(metrics_df.columns)) - set(merge_keys)
ia_df = ia_df.merge(
metrics_df.drop(columns=list(drop_cols) + ['Morph']).drop_duplicates(),
on=merge_keys,
how='left',
)
# rename linguistic feature columns
ia_df = ia_df.rename(columns=self.LINGUISTIC_FEATURE_RENAMES)
# initialize missing columns
ia_df, fix_df = self._initialize_missing_columns(ia_df, fix_df)
# add stratify column DE_RC
ia_df['DE_RC'] = ia_df['DE'].astype(str) + '_' + ia_df['RC'].astype(str)
fix_df['DE_RC'] = fix_df['DE'].astype(str) + '_' + fix_df['RC'].astype(str)
# map fixations to aois
word_aoi_df = self._read_word_aoi_df()
fix_df = self.map_to_aois(fix_df, word_aoi_df)
fix_df['NEXT_FIX_INTEREST_AREA_INDEX'] = fix_df[
'CURRENT_FIX_INTEREST_AREA_INDEX'
].shift(-1)
# prepare merge
merge_keys = self.data_args.groupby_columns + [
Fields.FIXATION_REPORT_IA_ID_COL_NAME
]
dup_cols = (set(fix_df.columns) & set(ia_df.columns)) - set(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.drop_duplicates(subset=merge_keys, keep='first'),
on=merge_keys,
how='left',
validate='many_to_one',
)
num_words = (
ia_df.groupby(self.data_args.groupby_columns)
.size()
.rename('num_of_words_in_trial')
)
enriched_fix_df = enriched_fix_df.merge(
num_words,
on=self.data_args.groupby_columns,
how='left',
)
enriched_fix_df['normalized_ID'] = enriched_fix_df[
'CURRENT_FIX_INTEREST_AREA_INDEX'
]
return enriched_fix_df, ia_df
def _read_word_aoi_df(self) -> pd.DataFrame:
stim_path = 'data/PoTeC/stimuli'
dfs = []
for file_name in os.listdir(stim_path):
if file_name.startswith('word_aoi_'):
file_path = os.path.join(stim_path, file_name)
tmp_df = pd.read_csv(file_path, delimiter='\t')
tmp_df['text_name'] = file_name.removesuffix('.tsv').split('_')[-1]
dfs.append(tmp_df)
return pd.concat(dfs, ignore_index=True)
def _load_and_merge_labels(self, df: pd.DataFrame) -> pd.DataFrame:
# participant data
label_df = pd.read_csv(
'data/PoTeC/labels/participant_data.tsv',
delimiter='\t',
).rename(columns={'reader_id': Fields.SUBJECT_ID})
merge_keys = [Fields.SUBJECT_ID]
drop_cols = (set(df.columns) & set(label_df.columns)) - set(merge_keys)
df = df.merge(
label_df.drop(columns=list(drop_cols)),
on=merge_keys,
validate='many_to_one',
)
# response accuracy data
text_spec_df = pd.read_csv(
'data/PoTeC/labels/participant_response_accuracy.tsv',
delimiter='\t',
).rename(
columns={
'reader_id': Fields.SUBJECT_ID,
'text_id': Fields.UNIQUE_PARAGRAPH_ID,
}
)
# drop existing accuracy columns if present
acc_cols = ['mean_acc_bq', 'acc_tq_1', 'acc_tq_2', 'acc_tq_3']
df = df.drop(columns=[col for col in acc_cols if col in df.columns])
merge_keys = [Fields.SUBJECT_ID, Fields.UNIQUE_PARAGRAPH_ID]
drop_cols = (set(df.columns) & set(text_spec_df.columns)) - set(merge_keys)
df = df.merge(
text_spec_df.drop(columns=list(drop_cols)),
on=merge_keys,
how='left',
validate='many_to_one',
)
# this is equivalent to everything correct
df['DE'] = (df['mean_acc_bq'] > 0.9).astype(int)
return df
def _expand_for_questions(self, df: pd.DataFrame) -> pd.DataFrame:
"""Expand dataframe to have separate rows for each question"""
dfs = []
question_indices = ['tq_1', 'tq_2', 'tq_3']
for q in tqdm(question_indices, desc='Expanding for questions'):
df_temp = df.copy()
df_temp['RC'] = df_temp[f'acc_{q}']
df_temp['question_index'] = q
df_temp['question'] = df_temp[q]
dfs.append(df_temp)
df = pd.concat(dfs, ignore_index=True)
df['unique_trial_id'] = (
df[Fields.SUBJECT_ID].astype(str)
+ '_'
+ df[Fields.UNIQUE_PARAGRAPH_ID].astype(str)
+ '_'
+ df['question_index']
)
return df
def dataset_specific_processing(
self, data_dict: dict[str, pd.DataFrame]
) -> dict[str, pd.DataFrame]:
"""PoTeC-specific processing pipeline"""
for data_type in [DataType.IA, DataType.FIXATIONS]:
logger.info(f'Processing {data_type} data')
if data_type not in data_dict or data_dict[data_type] is None:
continue
df = data_dict[data_type]
# exclude trials without comprehension scores
question_indices = ['tq_1', 'tq_2', 'tq_3']
acc_columns = [f'acc_{q}' for q in question_indices]
df = df.dropna(subset=acc_columns).copy()
# load and merge labels
df = self._load_and_merge_labels(df)
# merge stimuli data
stim_df = pd.read_csv(
'data/PoTeC/stimuli/stimuli.tsv', delimiter='\t'
).rename(
columns={
'reader_id': Fields.SUBJECT_ID,
'text_id': Fields.UNIQUE_PARAGRAPH_ID,
}
)
df = df.merge(
stim_df, on=[Fields.UNIQUE_PARAGRAPH_ID], validate='many_to_one'
)
# duplicate df for questions
df = self._expand_for_questions(df)
data_dict[data_type] = df
data_dict['fixations'], data_dict['ia'] = (
self.add_ia_report_features_to_fixation_data(
data_dict['ia'],
data_dict['fixations'],
)
)
for data_type in [DataType.IA, DataType.FIXATIONS]:
data_dict[data_type] = add_missing_features(
et_data=data_dict[data_type],
trial_groupby_columns=self.data_args.groupby_columns,
mode=data_type,
)
for data_type in data_dict.keys():
df = data_dict[data_type]
data_dict[data_type] = df.loc[:, ~df.columns.duplicated()]
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
data_dict = replace_missing_values(data_dict)
return data_dict
|