-
Notifications
You must be signed in to change notification settings - Fork 0
/
boto_wrap.py
430 lines (364 loc) · 16.7 KB
/
boto_wrap.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
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
import pickle
import boto
import pandas as pd
import json
import jsonschema
from collections import defaultdict
from copy import deepcopy
import boto.mturk.connection as tc
import boto.mturk.question as tq
from boto.mturk.qualification import PercentAssignmentsApprovedRequirement, Qualifications, Requirement
import requests
from annotation_schema import page_schema
"""
This module defines several functions used in the example mechanical-turk jupyter notebook.
Some will be more universally useful than others. For instance, when column names are specified they are likely unique
to my dataset/task.
"""
def load_book_info():
"""
loads information specific to my textbook dataset.
:return: textbook grouping and page ranges to use.
"""
with open('breakdowns.pkl', 'rb') as f:
book_breakdowns = pickle.load(f)
with open('pdfs/page_ranges.csv') as f:
ranges = f.readlines()
range_lookup = {line.split(' ')[0]:[int(num) for num in line.strip().split(' ')[1:]] for line in ranges}
return book_breakdowns, range_lookup
def form_hit_url(book_name, page_n):
"""
This generates the url that amt will use when creating a HIT.
:param book_name: book name base string
:param page_n: page_n int
:return: formatted url string
"""
book_name_no_ext = book_name.replace('.pdf', '_')
base_url = 'https://s3-us-west-2.amazonaws.com/ai2-vision-turk-data/textbook-annotation-test/build/index.html'
full_url = base_url + '?url={}{}.jpeg&id={}'.format(book_name_no_ext, page_n, page_n)
return full_url
def make_book_group_urls(book_groups, book_group, ranges):
"""
Makes urls for many pages and books in a particular group.
:param book_groups: book group definitions
:param book_group: group to process
:param ranges: page ranges to use
:return: list of formatted urls
"""
def get_start_end(start, end):
return start, end
group_urls = []
for tb in book_groups[book_group]:
start, end = get_start_end(*ranges[tb])
for page_n in range(start, end):
group_urls.append(form_hit_url(tb, page_n))
return group_urls
def build_hit_params(url, static_params):
"""
Dynamically builds some HIT params that will change based on the book/url
:param url: formatted url of page image on s3
:param static_params: Universal HIT params (set by user in notebook).
:return: complete HIT parameters.
"""
def build_qualifications():
"""
Creates a single qualification that workers have a > 95% acceptance rate.
:return: boto qualification obj.
"""
qualifications = Qualifications()
req1 = PercentAssignmentsApprovedRequirement(comparator="GreaterThan", integer_value="95")
qualifications.add(req1)
return qualifications
hit_params = deepcopy(static_params)
hit_params['qualifications'] = build_qualifications()
hit_params['questionform'] = tq.ExternalQuestion(url, static_params['frame_height'])
hit_params['reward'] = boto.mturk.price.Price(hit_params['amount'])
return hit_params
def create_single_hit(mturk_connection, url, static_hit_params):
"""
Creates a single HIT from a provided url
:param mturk_connection: active mturk connection established by user in the nb.
:param url: page url for the HIT
:param static_hit_params: User-defined global HIT params
:return: boto create hit return as a status check
"""
hit_params = build_hit_params(url, static_hit_params)
create_hit_result = mturk_connection.create_hit(
title=hit_params['title'],
description=hit_params['description'],
keywords=hit_params['keywords'],
question=hit_params['questionform'],
reward=hit_params['reward'],
max_assignments=hit_params['max_assignments'],
duration=hit_params['duration'],
qualifications=hit_params['qualifications'],
lifetime=hit_params['lifetime']
)
return create_hit_result
def create_hits_from_pages(mturk_connection, page_links, static_hit_params):
for url in page_links:
create_single_hit(mturk_connection, url, static_hit_params)
def delete_all_hits(mturk_connection):
"""
Permanently disables/ deletes all of the users active HITs.
:param mturk_connection: active mturk connection established by user in the nb.
:return:
"""
my_hits = list(mturk_connection.get_all_hits())
for hit in my_hits:
mturk_connection.disable_hit(hit.HITId)
def count_pages_in_df(any_result_df):
return len(pd.unique(any_result_df['page']))
def count_pages_with_cat(consensus_df, category):
return len(pd.unique(consensus_df[consensus_df['category'] == category]['page']))
def delete_some_hits(mturk_connection, hit_ids):
for hit in hit_ids.keys():
mturk_connection.disable_hit(hit)
def get_completed_hits(mturk_connection):
"""
Queries amt for all active user HITs.
:param mturk_connection: active mturk connection established by user in the nb.
:return: list of boto HIT result objects
"""
reviewable_hits = []
page_n = 1
hits_left = True
while hits_left:
hit_range = mturk_connection.get_reviewable_hits(page_size=100, page_number=page_n)
if not hit_range:
hits_left = False
break
reviewable_hits.extend(hit_range)
page_n += 1
return reviewable_hits
def get_assignments(mturk_connection, reviewable_hits, status=None):
"""
Retrieves individual assignments associated with the specified HITs.
:param mturk_connection: active mturk connection established by user in the nb.
:param reviewable_hits: HITs to review
:param status: HIT status to filter by.
:return: hit_id:assignment dict
"""
assignments = defaultdict(list)
for hit in reviewable_hits:
assignment = mturk_connection.get_assignments(hit.HITId, status=status)
assignments[hit.HITId].extend(assignment)
return assignments
def process_raw_hits(assignments_by_hit):
"""
Extracts assignment results from boto assignment objects in a more convienent form.
:param assignments_by_hit: dict of boto assignment objects.
:return: A nested dict with the box_id:assigned labels at the lowest level.
"""
mechanical_turk_results = defaultdict(list)
for hit_id, hit_assignments in assignments_by_hit.items():
for assignment in hit_assignments:
for answers in assignment.answers:
box_result = answers[1].fields[0]
box_json = json.loads(box_result)
for box in box_json:
box['worker_id'] = assignment.WorkerId
mechanical_turk_results[hit_id].append({
assignment.AssignmentId: {answers[0].fields[0]: box_json}}
)
return mechanical_turk_results
def accept_hits(mturk_connection, assignments_to_approve):
for assignment in assignments_to_approve:
if assignment.AssignmentStatus == 'Submitted':
mturk_connection.approve_assignment(assignment.AssignmentId)
else:
print assignment.AssignmentStatus
def match_workers_assignments(worker_list, worker_result_df):
"""
Creates a dataframe with results only from specified workers.
:param worker_list: workers to filter on
:param worker_result_df: all worker results
:return: results filtered by worker
"""
match_df = worker_result_df[worker_result_df['worker_id'].isin(worker_list)]
return pd.unique(match_df['assignment_id']).tolist(), pd.unique(match_df['worker_id']).tolist()
def reject_assignments(mturk_connection, workers_to_reject, worker_result_df):
feedback_message = """
Your HITs contained many incomplete or incorrect pages.
"""
assignments_to_reject, workers_rejected = match_workers_assignments(workers_to_reject, worker_result_df)
reject_count = len(assignments_to_reject)
worker_count = len(workers_rejected)
for assignment_id in assignments_to_reject:
try:
mturk_connection.reject_assignment(assignment_id, feedback_message)
except boto.mturk.connection.MTurkRequestError:
print 'assignment ' + str(assignment_id) + ' already accepted or rejected'
return reject_count, worker_count
def ban_bad_workers(mturk_connection, worker_ids):
for worker in worker_ids:
reason_for_block = """
Worker's submissions were largely incomplete.
"""
print 'blocking ' + str(worker)
mturk_connection.block_worker(worker, reason_for_block)
def get_assignment_statuses(assignment_results):
assignment_status = []
for hit_id, assignments in assignment_results.items():
for assignment in assignments:
assignment_status.append(assignment.AssignmentStatus)
status_series = pd.Series(assignment_status)
return status_series.value_counts()
def make_results_df(raw_hit_results):
"""
Creates a pandas dataframe from processed HIT results.
NOTE- the way I do this with pandas is not efficient, despite setting the new rows by loc.
Profiling this function pointed a finger at pandas, but I need to investigate further.
:param raw_hit_results: results dict processed using the process_raw_hits function above
:return: text-box level results in a pandas dataframe
"""
col_names = ['page', 'category', 'hit_id', 'assignment_id', 'box_id', 'worker_id']
results_df = pd.DataFrame(columns=col_names)
for hit_id, assignments in raw_hit_results.items():
for assignment in assignments:
for a_id, annotation in assignment.items():
for page, labeled_text in annotation.items():
for box in labeled_text:
results_df.loc[len(results_df)] = \
[page, box['category'], hit_id, a_id, box['id'], box['worker_id']]
return results_df
def make_question_results_df(raw_hit_results):
"""
similar to above with a new column for question group
"""
col_names = ['page', 'category', 'hit_id', 'assignment_id', 'box_id', 'worker_id', 'group_n']
results_df = pd.DataFrame(columns=col_names)
for hit_id, assignments in raw_hit_results.items():
for assignment in assignments:
for a_id, annotation in assignment.items():
for page, labeled_text in annotation.items():
for box in labeled_text:
if 'group_n' in box.keys():
group_n = box['group_n']
else:
group_n = 0
results_df.loc[len(results_df)] = \
[page, box['category'], hit_id, a_id, box['id'], box['worker_id'], str(group_n)]
return results_df
def make_consensus_df(results_df, no_consensus_flag):
"""
Computes consensus labels from turker responses.
:param results_df: result dataframe generated by the above function.
:param no_consensus_flag: value to fill in for boxes without consensus.
:return: consensus results
"""
grouped_by_page = results_df.groupby(['page', 'box_id'])
aggregated_df = grouped_by_page.agg(pd.DataFrame.mode)
aggregated_df.drop(['assignment_id', 'page', 'box_id', 'worker_id'], axis=1, inplace=True)
aggregated_df = aggregated_df.fillna(no_consensus_flag)
consensus_results_df = aggregated_df.reset_index()
consensus_results_df.drop('level_2', axis=1, inplace=True)
return consensus_results_df
def make_consensus_df_w_worker_id(combined_results_df, combined_consensus_results_df):
"""
Adds worker-level information to the consensus results
:param combined_results_df: results dataframe
:param combined_consensus_results_df: consensus results dataframe.
:return:
"""
consensus_with_worker_id_df = pd.DataFrame(columns=list(combined_consensus_results_df.columns) + ['worker_id', 'consensus_category'])
for hitbox_id, rows in combined_results_df.groupby(['hit_id', 'box_id']):
this_consensus_row = combined_consensus_results_df[
(combined_consensus_results_df['hit_id'] == hitbox_id[0]) & (combined_consensus_results_df['box_id'] == hitbox_id[1])]
new_rows = rows.copy()
new_rows['consensus_category'] = this_consensus_row['category'].values[0]
consensus_with_worker_id_df = consensus_with_worker_id_df.append(new_rows)
return consensus_with_worker_id_df
def form_annotation_url(page_name, anno_dir):
"""
generates annotation url to match page image url.
:param page_name:
:param anno_dir:
:return:
"""
base_path = '/Users/schwenk/wrk/notebooks/stb/ai2-vision-turk-data/textbook-annotation-test/'
file_path = base_path + anno_dir
return file_path + page_name.replace('jpeg', 'json')
def load_local_annotation(page_name, anno_dir):
"""
loads annotation from disk
:return: annotation json
"""
base_path = '/Users/schwenk/wrk/notebooks/stb/ai2-vision-turk-data/textbook-annotation-test/'
file_path = base_path + anno_dir + page_name.replace('jpeg', 'json')
try:
with open(file_path, 'r') as f:
local_annotations = json.load(f)
except IOError as e:
print e
local_annotations = None
return local_annotations
def process_annotation_results(anno_page_name, boxes, unannotated_page, annotations_folder, page_schema):
"""
read local annotations on disk and creates new annotation with consensus turk results
:param anno_page_name: page name
:param boxes: text boxes to process
:param unannotated_page: original annotation json
:param annotations_folder: destination dir to be written to
:param page_schema: page schema to validate against.
"""
question_cats = ['Multiple Choice',
'Fill-in-the-Blank',
'Short Answer',
'Discussion']
def update_box(result_row):
box_id = result_row['box_id']
category = result_row['category']
# group_n = result_row['group_n'] # this change is for the simpler question annotation task
group_n = 0
if box_id[0] == 'Q':
annotation_type = 'question'
unannotated_page[annotation_type][box_id]['category'] = category
unannotated_page[annotation_type][box_id]['group_n'] = group_n
elif category in question_cats:
old_annotation_type = 'text'
new_annotation_type = 'question'
new_id = box_id.replace('T', 'Q')
unannotated_page[new_annotation_type][new_id] = unannotated_page[old_annotation_type][box_id]
unannotated_page[new_annotation_type][new_id]['category'] = category
unannotated_page[new_annotation_type][new_id]['group_n'] = group_n
unannotated_page[new_annotation_type][new_id]['box_id'] = new_id
del unannotated_page[old_annotation_type][box_id]
unannotated_page[annotation_type][box_id.replace('Q', 'T')]['group_n'] = group_n
boxes.apply(update_box, axis=1)
file_path = annotations_folder + anno_page_name.replace('jpeg', 'json').replace("\\", "")
with open(file_path, 'wb') as f:
json.dump(unannotated_page, f)
return
def write_consensus_results(page_name, boxes, local_result_path, anno_dir):
"""
writes consensus results to disk.
"""
unaltered_annotations = load_local_annotation(page_name, anno_dir)
if unaltered_annotations:
process_annotation_results(page_name, boxes, unaltered_annotations, local_result_path, page_schema)
def write_results_df(aggregate_results_df, anno_dir, local_result_dir='newly-labeled-annotations/'):
"""
writes new annotation json to disk from a results dataframe
:param aggregate_results_df: dataframe to write
:param anno_dir: destination dir
:param local_result_dir: local original annotations to add to
"""
base_path = '/Users/schwenk/wrk/notebooks/stb/ai2-vision-turk-data/textbook-annotation-test/'
local_result_path = base_path + local_result_dir
for page, boxes in aggregate_results_df.groupby('page'):
write_consensus_results(page, boxes, local_result_path, anno_dir)
def review_results(pages_to_review, annotation_dir='newly-labeled-annotations/'):
"""
Sends a post request to a flask server running the annotation review tool specifying which pages to review.
:param pages_to_review: List of pages to review
:param annotation_dir: dir to load annotations from
:return: server response
"""
review_api_endpoint = 'http://localhost:8080/api/review'
payload = {'pages_to_review': str(pages_to_review), 'annotation_dir': annotation_dir}
headers = {'content-type': 'application/json'}
return requests.post(review_api_endpoint, data=json.dumps(payload), headers=headers)
def pickle_this(results_df, file_name):
with open(file_name, 'w') as f:
pickle.dump(results_df, f)