def test_hand_label_rating_events(self):
# This rating event isn't aligned with the task period below (it comes after).
# But we will still be able to associate it with that task created because
# with a list of hand labels that we pass in.
rating_event = create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 3, 0, 0),
event_type="Rating: 1",
url="http://url2.com",
user_id=2,
)
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
user_id=2,
task_index=4,
)
compute_location_ratings(labels=[
{'user_id': 2, 'task_index': 4, 'event_id': rating_event.id},
])
self.assertEqual(LocationRating.select().count(), 1)
rating = LocationRating.select().first()
self.assertEqual(rating.task_index, 4)
self.assertEqual(rating.hand_aligned, True)
def test_hand_label_rating_events(self):
# This rating event isn't aligned with the task period below (it comes after).
# But we will still be able to associate it with that task created because
# with a list of hand labels that we pass in.
rating_event = create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 3, 0, 0),
event_type="Rating: 1",
url="http://url2.com",
user_id=2,
)
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
user_id=2,
task_index=4,
)
compute_location_ratings(labels=[
{
'user_id': 2,
'task_index': 4,
'event_id': rating_event.id
},
])
self.assertEqual(LocationRating.select().count(), 1)
rating = LocationRating.select().first()
self.assertEqual(rating.task_index, 4)
self.assertEqual(rating.hand_aligned, True)
def add_location_rating(location, rating, comment):
"""
@summary: Add the location, rating and comment for the location
@param location : Query location
@type location: An object of Location class
@param rating: rating of the location as float
@type rating: float
@param comment: comment added for the location
@type comment: str
"""
loc_rating = LocationRating(loc, rating=rating, comment=comment)
loc_rating.save()
def main(page_types_json_filename, *args, **kwargs):
with open(page_types_json_filename) as page_types_file:
page_types = json.load(page_types_file)
# Only dump the most recently computed location ratings (ignore all others).
latest_compute_index = LocationRating.select(
fn.Max(LocationRating.compute_index)).scalar()
ratings = (LocationRating.select().where(
LocationRating.compute_index == latest_compute_index))
# Store a list of URLs for which labels are missing
urls_without_labels = set()
for rating in ratings:
# Get the domain name of where this rating happened
url_parsed = urlparse(rating.url)
domain = url_parsed.netloc.lstrip("www.")
# Fetch semantic labels for this URL
# Store missing URLs for non-pilot study participants.
# Currently, it's not important for us to be able to classify URLs for pilot participants.
unique_url = standardize_url(rating.url)
if unique_url not in page_types:
if rating.user_id > PILOT_MAX_USER_ID:
urls_without_labels.add(unique_url)
else:
page_type = page_types[unique_url]['main_type']
yield [[
rating.compute_index,
rating.user_id,
rating.task_index,
rating.concern_index,
rating.url,
domain,
page_type,
rating.rating,
rating.title,
rating.visit_date,
]]
# Print out a list of URLs for which labels were not found
for url in sorted(urls_without_labels):
logger.debug("No label found for URL: %s", url)
raise StopIteration
def test_by_default_associate_rating_with_latest_computed_task_periods(self):
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 0",
url="http://url1.com",
user_id=0,
)
# All three of these tasks have the same (matching) periods.
# But the second one was the latest one to be computed (compute_index=2)
create_task_period(
compute_index=0,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=1,
)
create_task_period(
compute_index=2,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=2,
)
create_task_period(
compute_index=1,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
)
compute_location_ratings()
rating = LocationRating.select()[0]
self.assertEqual(rating.task_index, 2)
def test_if_task_compute_index_specified_only_match_tasks_with_that_index(
self):
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 0",
url="http://url1.com",
user_id=0,
)
create_task_period(
compute_index=0,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=1,
)
create_task_period(
compute_index=2,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=2,
)
create_task_period(
compute_index=1,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
)
compute_location_ratings(task_compute_index=1)
rating = LocationRating.select()[0]
self.assertEqual(rating.task_index, 3)
def test_skip_ratings_that_couldnt_be_classified(self):
# The first and last events don't fall into a valid task period
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 11, 0, 0, 0),
event_type="Rating: 0",
url="http://url1.com",
user_id=0,
)
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 1",
url="http://url2.com",
user_id=0,
)
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 13, 0, 0, 0),
event_type="Rating: 2",
url="http://url2.com",
user_id=0,
)
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
user_id=0,
)
compute_location_ratings()
ratings = LocationRating.select()
self.assertEqual(ratings.count(), 1)
self.assertEqual(ratings.first().rating, 1)
def test_match_with_log_date_not_visit_date(self):
# Setup: create a rating event that occurred within a task
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
visit_date=datetime.datetime(9000, 1, 1, 1, 0, 0, 0),
event_type="Rating: 1",
url="http://url1.com",
user_id=0,
)
create_location_event(
log_date=datetime.datetime(9000, 1, 1, 1, 0, 0, 0),
visit_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 1",
url="http://url2.com",
user_id=0,
)
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
concern_index=5,
user_id=0,
)
# Test: a rating should be created for the URL visited with
# the index of the task and concern of the task period that was taking place at the time.
compute_location_ratings()
ratings = LocationRating.select()
self.assertEqual(ratings.count(), 1)
self.assertEqual(ratings.first().url, "http://url1.com")
def test_skip_ratings_that_couldnt_be_classified(self):
# The first and last events don't fall into a valid task period
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 11, 0, 0, 0),
event_type="Rating: 0",
url="http://url1.com",
user_id=0,
)
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 1",
url="http://url2.com",
user_id=0,
)
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 13, 0, 0, 0),
event_type="Rating: 2",
url="http://url2.com",
user_id=0,
)
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
user_id=0,
)
compute_location_ratings()
ratings = LocationRating.select()
self.assertEqual(ratings.count(), 1)
self.assertEqual(ratings.first().rating, 1)
def test_if_task_compute_index_specified_only_match_tasks_with_that_index(self):
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 0",
url="http://url1.com",
user_id=0,
)
create_task_period(
compute_index=0,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=1,
)
create_task_period(
compute_index=2,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=2,
)
create_task_period(
compute_index=1,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
)
compute_location_ratings(task_compute_index=1)
rating = LocationRating.select()[0]
self.assertEqual(rating.task_index, 3)
def test_by_default_associate_rating_with_latest_computed_task_periods(
self):
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 0",
url="http://url1.com",
user_id=0,
)
# All three of these tasks have the same (matching) periods.
# But the second one was the latest one to be computed (compute_index=2)
create_task_period(
compute_index=0,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=1,
)
create_task_period(
compute_index=2,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=2,
)
create_task_period(
compute_index=1,
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
)
compute_location_ratings()
rating = LocationRating.select()[0]
self.assertEqual(rating.task_index, 2)
def test_match_with_log_date_not_visit_date(self):
# Setup: create a rating event that occurred within a task
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
visit_date=datetime.datetime(9000, 1, 1, 1, 0, 0, 0),
event_type="Rating: 1",
url="http://url1.com",
user_id=0,
)
create_location_event(
log_date=datetime.datetime(9000, 1, 1, 1, 0, 0, 0),
visit_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 1",
url="http://url2.com",
user_id=0,
)
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
concern_index=5,
user_id=0,
)
# Test: a rating should be created for the URL visited with
# the index of the task and concern of the task period that was taking place at the time.
compute_location_ratings()
ratings = LocationRating.select()
self.assertEqual(ratings.count(), 1)
self.assertEqual(ratings.first().url, "http://url1.com")
def create_location_rating(compute_index, task_compute_index, event, rating, labels):
''' Returns True if this rating could be matched to an existing task, False otherwise. '''
# Check for hand-written task index labels for this event
matching_labels = filter(lambda l: l['event_id'] == event.id, labels)
if len(matching_labels) > 0:
task_index = matching_labels[0]['task_index']
task_periods = (
TaskPeriod.select()
.where(
TaskPeriod.compute_index == task_compute_index,
TaskPeriod.task_index == task_index,
TaskPeriod.user_id == event.user_id,
)
)
hand_aligned = True
# If a hand-written label wasn't found, search for a task that this rating could have
# occurred within. If we successfully find a task, then save the rating event.
else:
task_periods = (
TaskPeriod.select()
.where(
TaskPeriod.compute_index == task_compute_index,
TaskPeriod.user_id == event.user_id,
TaskPeriod.start < event.log_date,
TaskPeriod.end > event.log_date,
)
)
hand_aligned = False
# If a matching task has been found, then save the rating alongside that task.
if task_periods.count() > 0:
task_period = task_periods[0]
LocationRating.create(
compute_index=compute_index,
user_id=event.user_id,
task_index=task_period.task_index,
concern_index=task_period.concern_index,
url=event.url,
rating=rating,
title=event.title,
visit_date=event.visit_date,
hand_aligned=hand_aligned,
)
return (task_periods.count() > 0)
def create_location_rating(compute_index, task_compute_index, event, rating,
labels):
''' Returns True if this rating could be matched to an existing task, False otherwise. '''
# Check for hand-written task index labels for this event
matching_labels = filter(lambda l: l['event_id'] == event.id, labels)
if len(matching_labels) > 0:
task_index = matching_labels[0]['task_index']
task_periods = (TaskPeriod.select().where(
TaskPeriod.compute_index == task_compute_index,
TaskPeriod.task_index == task_index,
TaskPeriod.user_id == event.user_id,
))
hand_aligned = True
# If a hand-written label wasn't found, search for a task that this rating could have
# occurred within. If we successfully find a task, then save the rating event.
else:
task_periods = (TaskPeriod.select().where(
TaskPeriod.compute_index == task_compute_index,
TaskPeriod.user_id == event.user_id,
TaskPeriod.start < event.log_date,
TaskPeriod.end > event.log_date,
))
hand_aligned = False
# If a matching task has been found, then save the rating alongside that task.
if task_periods.count() > 0:
task_period = task_periods[0]
LocationRating.create(
compute_index=compute_index,
user_id=event.user_id,
task_index=task_period.task_index,
concern_index=task_period.concern_index,
url=event.url,
rating=rating,
title=event.title,
visit_date=event.visit_date,
hand_aligned=hand_aligned,
)
return (task_periods.count() > 0)
def main(page_types_json_filename, *args, **kwargs):
with open(page_types_json_filename) as page_types_file:
page_types = json.load(page_types_file)
# Only dump the most recently computed location ratings (ignore all others).
latest_compute_index = LocationRating.select(fn.Max(LocationRating.compute_index)).scalar()
ratings = (
LocationRating
.select()
.where(
LocationRating.compute_index == latest_compute_index
)
)
# Store a list of URLs for which labels are missing
urls_without_labels = set()
for rating in ratings:
# Get the domain name of where this rating happened
url_parsed = urlparse(rating.url)
domain = url_parsed.netloc.lstrip("www.")
# Fetch semantic labels for this URL
# Store missing URLs for non-pilot study participants.
# Currently, it's not important for us to be able to classify URLs for pilot participants.
unique_url = standardize_url(rating.url)
if unique_url not in page_types:
if rating.user_id > PILOT_MAX_USER_ID:
urls_without_labels.add(unique_url)
else:
page_type = page_types[unique_url]['main_type']
yield [[
rating.compute_index,
rating.user_id,
rating.task_index,
rating.concern_index,
rating.url,
domain,
page_type,
rating.rating,
rating.title,
rating.visit_date,
]]
# Print out a list of URLs for which labels were not found
for url in sorted(urls_without_labels):
logger.debug("No label found for URL: %s", url)
raise StopIteration
def test_dont_make_rating_for_non_rating_event(self):
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Tab opened", # this event is not a rating and shouldn't be read as one
url="http://url1.com",
)
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
concern_index=5,
)
compute_location_ratings()
self.assertEqual(LocationRating.select().count(), 0)
def test_dont_make_rating_for_non_rating_event(self):
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type=
"Tab opened", # this event is not a rating and shouldn't be read as one
url="http://url1.com",
)
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
concern_index=5,
)
compute_location_ratings()
self.assertEqual(LocationRating.select().count(), 0)
def compute_location_ratings(labels=HAND_LABELED_EVENTS,
task_compute_index=None):
# Create a new index for this computation
last_compute_index = LocationRating.select(
fn.Max(LocationRating.compute_index)).scalar() or 0
compute_index = last_compute_index + 1
# Determine what will be the compute index of the task periods that ratings are matched to.
# This will become the latest compute index if it hasn't been specified.
if task_compute_index is None:
task_compute_index = TaskPeriod.select(fn.Max(
TaskPeriod.compute_index)).scalar()
# Create a list to hold all ratings that couldn't be matched to a task period.
# At the end, we want to return these, in case it's important for the caller to know
# which events we couldn't create rating records for.
unmatched_ratings = []
for event in LocationEvent.select():
# Check to see whether this is a rating event
rating_match = re.match("^Rating: (\d)+$", event.event_type)
if rating_match:
# If this is a rating event, extract the rating
rating = int(rating_match.group(1))
rating_created = create_location_rating(
compute_index=compute_index,
task_compute_index=task_compute_index,
event=event,
rating=rating,
labels=labels,
)
# If a rating wasn't created, this probably couldn't be matched to a task.
# Save a record of which event failed to be matched to a task and which user
# this event happened for.
if not rating_created:
unmatched_ratings.append({
'user_id': event.user_id,
'event_id': event.id,
})
return unmatched_ratings
def compute_location_ratings(labels=HAND_LABELED_EVENTS, task_compute_index=None):
# Create a new index for this computation
last_compute_index = LocationRating.select(fn.Max(LocationRating.compute_index)).scalar() or 0
compute_index = last_compute_index + 1
# Determine what will be the compute index of the task periods that ratings are matched to.
# This will become the latest compute index if it hasn't been specified.
if task_compute_index is None:
task_compute_index = TaskPeriod.select(fn.Max(TaskPeriod.compute_index)).scalar()
# Create a list to hold all ratings that couldn't be matched to a task period.
# At the end, we want to return these, in case it's important for the caller to know
# which events we couldn't create rating records for.
unmatched_ratings = []
for event in LocationEvent.select():
# Check to see whether this is a rating event
rating_match = re.match("^Rating: (\d)+$", event.event_type)
if rating_match:
# If this is a rating event, extract the rating
rating = int(rating_match.group(1))
rating_created = create_location_rating(
compute_index=compute_index,
task_compute_index=task_compute_index,
event=event,
rating=rating,
labels=labels,
)
# If a rating wasn't created, this probably couldn't be matched to a task.
# Save a record of which event failed to be matched to a task and which user
# this event happened for.
if not rating_created:
unmatched_ratings.append({
'user_id': event.user_id,
'event_id': event.id,
})
return unmatched_ratings
def test_location_event_must_match_task_period_and_user_id(self):
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 0",
url="http://url1.com",
user_id=0,
)
# For the task periods below, the `task_index` parameter serves no function,
# except to vary between all periods so that we know which task the rating
# is associated with after it has been extracted.
# This task is a mismatch because it's for the wrong user.
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=2,
user_id=1,
)
# This task is a match, with the right timing and the right user.
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
user_id=0,
)
# This task is a mismatch, with the wrong timing but the right user.
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 4, 0, 0),
task_index=4,
user_id=0,
)
compute_location_ratings()
ratings = LocationRating.select()
self.assertEqual(ratings.count(), 1)
rating = ratings[0]
self.assertEqual(rating.user_id, 0)
self.assertEqual(rating.task_index, 3)
def test_location_event_must_match_task_period_and_user_id(self):
create_location_event(
log_date=datetime.datetime(2000, 1, 1, 12, 0, 1, 0),
event_type="Rating: 0",
url="http://url1.com",
user_id=0,
)
# For the task periods below, the `task_index` parameter serves no function,
# except to vary between all periods so that we know which task the rating
# is associated with after it has been extracted.
# This task is a mismatch because it's for the wrong user.
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=2,
user_id=1,
)
# This task is a match, with the right timing and the right user.
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 0, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
task_index=3,
user_id=0,
)
# This task is a mismatch, with the wrong timing but the right user.
create_task_period(
start=datetime.datetime(2000, 1, 1, 12, 2, 0, 0),
end=datetime.datetime(2000, 1, 1, 12, 4, 0, 0),
task_index=4,
user_id=0,
)
compute_location_ratings()
ratings = LocationRating.select()
self.assertEqual(ratings.count(), 1)
rating = ratings[0]
self.assertEqual(rating.user_id, 0)
self.assertEqual(rating.task_index, 3)