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 | def compute_dataset_profile(
dataset_name: str,
mode: DataType,
curr_df: pd.DataFrame,
output_path: Path,
data_args: DataArgs,
):
"""
Computes and updates per-dataset profile stats in a shared CSV.
"""
logger.info(f'---- Stats for: {dataset_name} ({mode}) ----')
stats = {}
item_col = data_args.unique_item_column
participant_col = data_args.subject_column
task_and_labels = data_args.tasks
additional_info = {
key: getattr(data_args, key, None)
for key in ['text_source', 'text_language', 'text_domain', 'text_type']
if hasattr(data_args, key)
}
if mode == DataType.IA:
stats = {
'n_participants': curr_df[participant_col].nunique(),
'n_items': curr_df[item_col].nunique(),
'n_words': curr_df.shape[0],
'n_trials': curr_df['unique_trial_id'].nunique(),
'n_words_per_participant': format_stats(
curr_df.groupby(participant_col).size()
),
'n_words_per_item': format_stats(curr_df.groupby(item_col).size()),
}
trial_level = curr_df.drop_duplicates(subset=['unique_trial_id'])
# label stats
if dataset_name == 'OneStop':
stats['n_words_corpus'] = 19428 # hardcoded for OneStop (based on Adv)
elif dataset_name == 'CopCo':
data_parags = curr_df[
['part', 'unique_paragraph_id', 'paragraph']
].drop_duplicates()
data_parags['n_words'] = data_parags['paragraph'].apply(
lambda x: len((x).split())
)
stats['n_words_corpus'] = int(data_parags['n_words'].sum())
df_to_save = data_parags[
['n_words', 'part', 'unique_paragraph_id', 'paragraph']
].sort_values(by=['part', 'unique_paragraph_id'])
df_to_save.to_csv(STATS_FOLDER / 'CopCo_paragraphs.csv', index=False)
logger.info(f'n_words calculated from {data_parags.shape[0]} paragraphs.')
elif dataset_name == 'PoTeC':
data_parags = (
curr_df[['unique_paragraph_id', 'paragraph']]
.drop_duplicates()
.sort_values(by='unique_paragraph_id')
)
data_parags['n_words'] = data_parags['paragraph'].apply(
lambda x: len((x).split())
)
stats['n_words_corpus'] = int(data_parags['n_words'].sum())
logger.info(f'Total n_words: {stats["n_words_corpus"]}')
df_to_save = data_parags[
['n_words', 'unique_paragraph_id', 'paragraph']
].sort_values(by='unique_paragraph_id')
df_to_save.to_csv(
STATS_FOLDER / f'{dataset_name}_paragraphs.csv', index=False
)
logger.info(f'n_words calculated from {data_parags.shape[0]} paragraphs.')
elif dataset_name == 'SBSAT':
data_parags = (
curr_df[['unique_paragraph_id', 'paragraph']]
.sort_values(by='unique_paragraph_id')
.drop_duplicates()
.sort_values(by='unique_paragraph_id')
)
data_parags['n_words'] = data_parags['paragraph'].apply(
lambda x: len((x).split())
)
# TODO: fix bug in SBSAT data
logger.warning(
'using different caclulation because of BUG in SBSAT paragraphs'
)
grouped = (
data_parags.groupby('unique_paragraph_id')[
['unique_paragraph_id', 'n_words']
]
.max()
.reset_index(drop=True)
)
stats['n_words_corpus'] = int(grouped['n_words'].sum())
logger.info(f'Total n_words: {stats["n_words_corpus"]}')
df_to_save = data_parags[
['n_words', 'unique_paragraph_id', 'paragraph']
].sort_values(by='unique_paragraph_id')
df_to_save.to_csv(
STATS_FOLDER / f'{dataset_name}_paragraphs.csv', index=False
)
logger.info(f'n_words calculated from {data_parags.shape[0]} paragraphs.')
elif dataset_name == 'MECOL2':
data_parags = (
curr_df[['paragraph', 'itemid', 'unique_paragraph_id']]
.drop_duplicates()
.sort_values(by='itemid')
.reset_index(drop=True)
)
data_parags['n_words'] = data_parags['paragraph'].apply(
lambda x: len((x).split())
)
# TODO: fix bug in MECO data
logger.warning(
'using different caclulation because of BUG in MECO paragraphs'
)
grouped = (
data_parags.groupby('itemid')[['itemid', 'n_words']]
.min()
.reset_index(drop=True)
)
stats['n_words_corpus'] = int(grouped['n_words'].sum())
logger.info(f'Total n_words: {stats["n_words_corpus"]}')
df_to_save = data_parags[
['n_words', 'itemid', 'unique_paragraph_id', 'paragraph']
].sort_values(by='itemid')
df_to_save.to_csv(
STATS_FOLDER / f'{dataset_name}_paragraphs.csv', index=False
)
logger.info(f'n_words calculated from {data_parags.shape[0]} paragraphs.')
elif dataset_name == 'IITBHGC':
data_parags = (
curr_df[['unique_paragraph_id', 'paragraph']]
.drop_duplicates()
.sort_values(by='unique_paragraph_id')
)
data_parags['n_words'] = data_parags['paragraph'].apply(
lambda x: len((x).split())
)
stats['n_words_corpus'] = int(data_parags['n_words'].sum())
logger.info(f'Total n_words: {stats["n_words_corpus"]}')
df_to_save = data_parags[
['n_words', 'unique_paragraph_id', 'paragraph']
].sort_values(by='unique_paragraph_id')
df_to_save.to_csv(
STATS_FOLDER / f'{dataset_name}_paragraphs.csv', index=False
)
logger.info(f'n_words calculated from {data_parags.shape[0]} paragraphs.')
else:
logger.info(f'Unknown dataset: {dataset_name}')
stats['n_words_corpus'] = '???'
for task, label_col in task_and_labels.items():
stats[f'{task}_col'] = label_col
# if label_col is boolean, convert to int
if pd.api.types.is_bool_dtype(curr_df[label_col]):
curr_df[label_col] = curr_df[label_col].astype(int)
# calc mean if not categorical
if pd.api.types.is_numeric_dtype(curr_df[label_col]):
stats[f'{task}_overall'] = format_stats(
trial_level[label_col], rounding=2
)
stats[f'{task}_per_participant'] = format_stats(
trial_level.groupby(participant_col)[label_col].mean(), rounding=2
)
stats[f'{task}_per_item'] = format_stats(
trial_level.groupby(item_col)[label_col].mean(), rounding=2
)
# label distribution if n unique values is small
if trial_level[label_col].nunique() <= 15:
label_distribution = (
trial_level[label_col].value_counts(normalize=True) * 100
).round(1)
stats[f'{task}_distribution'] = label_distribution.to_dict()
elif mode == DataType.FIXATIONS:
stats = {
'n_fix': curr_df.shape[0],
'n_fix_per_trial': format_stats(curr_df.groupby('unique_trial_id').size()),
'n_fix_per_participant': format_stats(
curr_df.groupby(participant_col).size()
),
'n_fix_per_item': format_stats(curr_df.groupby(item_col).size()),
}
elif mode == DataType.TRIAL_LEVEL:
pass # TODO: add?
# save to stats all fields in additional_info_dataset except task_and_labels
for key, val in additional_info.items():
if key != 'task_and_labels':
stats[key] = val
# Load or create the profile DataFrame
if output_path.exists():
profile_df = pd.read_csv(output_path, index_col=0)
else:
profile_df = pd.DataFrame()
# Initialize profile_df if empty
if profile_df.empty:
profile_df = pd.DataFrame(index=stats.keys(), columns=[data_args.dataset_name])
else:
for key in stats.keys():
if key not in profile_df.index:
profile_df.loc[key] = pd.Series()
# Assign each stat
for key, val in stats.items():
profile_df.at[key, data_args.dataset_name] = val
profile_df.sort_index(inplace=True)
output_path.parent.mkdir(parents=True, exist_ok=True)
profile_df.to_csv(output_path)
logger.info(
f'Updated profile for {data_args.dataset_name} ({mode}) → {output_path}'
)
|