def validate_times(srt_content, srt_issues):
times = re.findall("([0-9:,]+) --> ([0-9:,]+)\r\n", srt_content, re.S | re.M)
parse_time = lambda str: datetime.datetime.strptime(str, "%H:%M:%S,%f")
for i in range(len(times)):
try:
between_subtitle_time = datediff(
parse_time(times[i][0]), parse_time(times[i - 1][1] if i > 0 else "00:00:00,000")
)
within_subtitle_time = datediff(parse_time(times[i][1]), parse_time(times[i][0]))
if between_subtitle_time > 60.0:
srt_issues.append("Between-subtitle gap of %5.2f seconds" % between_subtitle_time)
if within_subtitle_time > 60.0:
srt_issues.append("Within-subtitle duration of %5.2f seconds" % within_subtitle_time)
elif within_subtitle_time == 0.0:
logging.debug("Subtitle flies by too fast (%s --> %s)." % times[i])
# print "Start: %s\tB: %5.2f\tW: %5.2f" % (parse_time(times[i][0]), between_subtitle_time, within_subtitle_time)
except Exception as e:
if not times[i][1].startswith("99:59:59"):
srt_issues.append("Error checking times: %s" % e)
else:
if len(times) - i > 1 and len(times) - i - 1 > len(times) / 10.0:
if i == 0:
srt_issues.append("No subtitles have a valid starting point.")
else:
logging.debug(
"Hit end of movie, but %d (of %d) subtitle(s) remain in the queue."
% (len(times) - i - 1, len(times))
)
break
def test_sign(self):
"""
Test the diff in both directions, validate they are the same and of opposite signs.
"""
d1 = datetime.datetime(2000, 1, 1, 0, 0, 0, 0)
d2 = datetime.datetime(2000, 1, 1, 0, 0, 0, 1) # test 4 years apart
self.assertTrue(datediff(d1,d2) < 0, "First date earlier than the second returns negative.")
self.assertTrue(datediff(d2,d1) > 0, "Second date earlier than the first returns positive.")
self.assertTrue(datediff(d2,d2) == 0, "First date equals the second returns 0.")
def add_to_summary(sender, **kwargs):
assert UserLog.is_enabled(), "We shouldn't be saving unless UserLog is enabled."
instance = kwargs["instance"]
if not instance.start_datetime:
raise ValidationError("start_datetime cannot be None")
if instance.last_active_datetime and instance.start_datetime > instance.last_active_datetime:
raise ValidationError("UserLog date consistency check for start_datetime and last_active_datetime")
if instance.end_datetime and not instance.total_seconds:
# Compute total_seconds, save to summary
# Note: only supports setting end_datetime once!
instance.full_clean()
# The top computation is more lenient: user activity is just time logged in, literally.
# The bottom computation is more strict: user activity is from start until the last "action"
# recorded--in the current case, that means from login until the last moment an exercise or
# video log was updated.
#instance.total_seconds = datediff(instance.end_datetime, instance.start_datetime, units="seconds")
instance.total_seconds = 0 if not instance.last_active_datetime else datediff(instance.last_active_datetime, instance.start_datetime, units="seconds")
# Confirm the result (output info first for easier debugging)
if instance.total_seconds < 0:
raise ValidationError("Total learning time should always be non-negative.")
logging.debug("%s: total time (%d): %d seconds" % (instance.user.username, instance.activity_type, instance.total_seconds))
# Save only completed log items to the UserLogSummary
UserLogSummary.add_log_to_summary(instance)
def handle(self, *args, **options):
old_map = os.path.exists(
DUBBED_VIDEOS_MAPPING_FILEPATH) and copy.deepcopy(
get_dubbed_video_map()) or {} # for comparison purposes
cache_filepath = options["cache_filepath"] or os.path.join(
settings.MEDIA_ROOT, 'khan_dubbed_videos.csv')
max_cache_age = (not options["force"]
and options["max_cache_age"]) or 0.0
if os.path.exists(cache_filepath) and datediff(
datetime.datetime.now(),
datetime.datetime.fromtimestamp(
os.path.getctime(cache_filepath)),
units="days") <= max_cache_age:
# Use cached data to generate the video map
csv_data = open(cache_filepath, "r").read()
else:
csv_data = download_ka_dubbed_video_mappings(
cache_filepath=cache_filepath)
# Use cached data to generate the video map
raw_map = generate_dubbed_video_mappings(csv_data=csv_data)
# Now we've built the map. Save it.
ensure_dir(os.path.dirname(DUBBED_VIDEOS_MAPPING_FILEPATH))
logging.info("Saving data to %s" % DUBBED_VIDEOS_MAPPING_FILEPATH)
with open(DUBBED_VIDEOS_MAPPING_FILEPATH, "w") as fp:
json.dump(raw_map, fp)
new_map = get_dubbed_video_map(reload=True)
# Now tell the user about what changed.
added_languages = set(new_map.keys()) - set(old_map.keys())
removed_languages = set(old_map.keys()) - set(new_map.keys())
if added_languages or removed_languages:
logging.info(
"*** Added support for %2d languages; removed support for %2d languages. ***"
% (len(added_languages), len(removed_languages)))
for lang_code in sorted(
list(set(new_map.keys()).union(set(old_map.keys())))):
added_videos = set(new_map[lang_code].keys()) - set(
old_map[lang_code].keys())
removed_videos = set(old_map[lang_code].keys()) - set(
new_map[lang_code].keys())
shared_keys = set(new_map[lang_code].keys()).intersection(
set(old_map[lang_code].keys()))
changed_videos = [
vid for vid in shared_keys
if old_map[lang_code][vid] != new_map[lang_code][vid]
]
logging.info(
"\t%5s: Added %d videos, removed %3d videos, changed %3d videos."
% (lang_code, len(added_videos), len(removed_videos),
len(changed_videos)))
logging.info("Done.")
def handle(self, *args, **options):
# Get the CSV data, either from a recent cache_file
# or from the internet
cache_dir = settings.MEDIA_ROOT
cache_file = os.path.join(cache_dir, "dubbed_videos.csv")
if not options["force"] and os.path.exists(cache_file) and datediff(datetime.datetime.now(), datetime.datetime.fromtimestamp(os.path.getctime(cache_file)), units="days") <= 14.0:
# Use cached data to generate the video map
csv_data = open(cache_file, "r").read()
(video_map, _) = generate_dubbed_video_mappings(csv_data=csv_data)
else:
# Use cached data to generate the video map
(video_map, csv_data) = generate_dubbed_video_mappings()
try:
ensure_dir(cache_dir)
with open(cache_file, "w") as fp:
fp.write(csv_data)
except Exception as e:
logging.error("Failed to make a local cache of the CSV data: %s" % e)
# Now we've built the map. Save it.
out_file = DUBBED_VIDEOS_MAPPING_FILEPATH
ensure_dir(os.path.dirname(out_file))
logging.info("Saving data to %s" % out_file)
with open(out_file, "w") as fp:
json.dump(video_map, fp)
logging.info("Done.")
def download_khan_data(url, debug_cache_file=None, debug_cache_dir=khanload.KHANLOAD_CACHE_DIR):
"""Download data from the given url.
In DEBUG mode, these downloads are slow. So for the sake of faster iteration,
save the download to disk and re-serve it up again, rather than download again,
if the file is less than a day old.
"""
# Get the filename
if not debug_cache_file:
debug_cache_file = url.split("/")[-1] + ".json"
# Create a directory to store these cached json files
if not os.path.exists(debug_cache_dir):
os.mkdir(debug_cache_dir)
debug_cache_file = os.path.join(debug_cache_dir, debug_cache_file)
# Use the cache file if:
# a) We're in DEBUG mode
# b) The debug cache file exists
# c) It's less than 7 days old.
if settings.DEBUG and os.path.exists(debug_cache_file) and datediff(datetime.datetime.now(), datetime.datetime.fromtimestamp(os.path.getctime(debug_cache_file)), units="days") <= 1E6:
# Slow to debug, so keep a local cache in the debug case only.
#sys.stdout.write("Using cached file: %s\n" % debug_cache_file)
with open(debug_cache_file, "r") as fp:
data = json.load(fp)
else:
sys.stdout.write("Downloading data from %s..." % url)
sys.stdout.flush()
data = json.loads(requests.get(url).content)
sys.stdout.write("done.\n")
# In DEBUG mode, store the debug cache file.
if settings.DEBUG:
with open(debug_cache_file, "w") as fh:
fh.write(json.dumps(data))
return data
def test_cancel_progress(self):
# Create the object
progress_log = UpdateProgressLog(process_name="test_process", total_stages=1)
progress_log.save()
# Complete the process
progress_log.cancel_progress()
self.assertTrue(abs(datediff(progress_log.end_time, datetime.datetime.now())) < 10, "end time is within 10 seconds")
self.assertEqual(progress_log.completed, False, "completed is False")
def download_khan_data(url,
debug_cache_file=None,
debug_cache_dir=KHANLOAD_CACHE_DIR):
"""Download data from the given url.
In DEBUG mode, these downloads are slow. So for the sake of faster iteration,
save the download to disk and re-serve it up again, rather than download again,
if the file is less than a day old.
"""
# Get the filename
if not debug_cache_file:
debug_cache_file = url.split("/")[-1] + ".json"
# Create a directory to store these cached json files
if not os.path.exists(debug_cache_dir):
os.mkdir(debug_cache_dir)
debug_cache_file = os.path.join(debug_cache_dir, debug_cache_file)
data = None
# Use the cache file if:
# a) We're in DEBUG mode
# b) The debug cache file exists
# c) It's less than 7 days old.
if settings.DEBUG and os.path.exists(debug_cache_file) and datediff(
datetime.datetime.now(),
datetime.datetime.fromtimestamp(
os.path.getctime(debug_cache_file)),
units="days") <= 1E6:
# Slow to debug, so keep a local cache in the debug case only.
#sys.stdout.write("Using cached file: %s\n" % debug_cache_file)
try:
with open(debug_cache_file, "r") as fp:
data = json.load(fp)
except Exception as e:
sys.stderr.write("Error loading cached document %s: %s\n" %
(debug_cache_file, e))
if data is None: # Failed to get a cached copy
sys.stdout.write("Downloading data from %s..." % url)
sys.stdout.flush()
try:
response = requests.get(url)
response.raise_for_status()
data = json.loads(response.content)
sys.stdout.write("done.\n")
except requests.HTTPError as e:
sys.stderr.write("Error downloading %s: %s\n" % (url, e))
return None
# In DEBUG mode, store the debug cache file.
if settings.DEBUG:
with open(debug_cache_file, "w") as fh:
fh.write(json.dumps(data))
return data
def test_completion(self):
# Create the object
progress_log = UpdateProgressLog(process_name="test_process", total_stages=1)
progress_log.save()
# Complete the process
progress_log.mark_as_completed()
self.assertTrue(abs(datediff(progress_log.end_time, datetime.datetime.now())) < 10, "end time is within 10 seconds")
self.assertEqual(progress_log.stage_percent, 1., "stage_percent==1")
self.assertEqual(progress_log.process_percent, 1., "proces_percent==1")
self.assertEqual(progress_log.completed, True, "completed is False")
def test_cancel_progress(self):
# Create the object
progress_log = UpdateProgressLog(process_name="test_process",
total_stages=1)
progress_log.save()
# Complete the process
progress_log.cancel_progress()
self.assertTrue(
abs(datediff(progress_log.end_time, datetime.datetime.now())) < 10,
"end time is within 10 seconds")
self.assertEqual(progress_log.completed, False, "completed is False")
def test_completion(self):
# Create the object
progress_log = UpdateProgressLog(process_name="test_process",
total_stages=1)
progress_log.save()
# Complete the process
progress_log.mark_as_completed()
self.assertTrue(
abs(datediff(progress_log.end_time, datetime.datetime.now())) < 10,
"end time is within 10 seconds")
self.assertEqual(progress_log.stage_percent, 1., "stage_percent==1")
self.assertEqual(progress_log.process_percent, 1., "proces_percent==1")
self.assertEqual(progress_log.completed, True, "completed is False")
def handle(self, *args, **options):
old_map = os.path.exists(DUBBED_VIDEOS_MAPPING_FILEPATH) and copy.deepcopy(get_dubbed_video_map()) or {} # for comparison purposes
cache_filepath = options["cache_filepath"] or os.path.join(settings.MEDIA_ROOT, 'khan_dubbed_videos.csv')
max_cache_age = (not options["force"] and options["max_cache_age"]) or 0.0
if os.path.exists(cache_filepath) and datediff(datetime.datetime.now(), datetime.datetime.fromtimestamp(os.path.getctime(cache_filepath)), units="days") <= max_cache_age:
# Use cached data to generate the video map
csv_data = open(cache_filepath, "r").read()
else:
csv_data = download_ka_dubbed_video_mappings(cache_filepath=cache_filepath)
# Use cached data to generate the video map
raw_map = generate_dubbed_video_mappings(csv_data=csv_data)
# Now we've built the map. Save it.
ensure_dir(os.path.dirname(DUBBED_VIDEOS_MAPPING_FILEPATH))
logging.info("Saving data to %s" % DUBBED_VIDEOS_MAPPING_FILEPATH)
with open(DUBBED_VIDEOS_MAPPING_FILEPATH, "w") as fp:
json.dump(raw_map, fp)
new_map = get_dubbed_video_map(reload=True)
# Now tell the user about what changed.
added_languages = set(new_map.keys()) - set(old_map.keys())
removed_languages = set(old_map.keys()) - set(new_map.keys())
if added_languages or removed_languages:
logging.info("*** Added support for %2d languages; removed support for %2d languages. ***" % (len(added_languages), len(removed_languages)))
for lang_code in sorted(list(set(new_map.keys()).union(set(old_map.keys())))):
added_videos = set(new_map[lang_code].keys()) - set(old_map[lang_code].keys())
removed_videos = set(old_map[lang_code].keys()) - set(new_map[lang_code].keys())
shared_keys = set(new_map[lang_code].keys()).intersection(set(old_map[lang_code].keys()))
changed_videos = [vid for vid in shared_keys if old_map[lang_code][vid] != new_map[lang_code][vid]]
logging.info("\t%5s: Added %d videos, removed %3d videos, changed %3d videos." % (lang_code, len(added_videos), len(removed_videos), len(changed_videos)))
logging.info("Done.")
def generate_fake_video_logs(facility_user=None,
topics=topics,
start_date=datetime.datetime.now() -
datetime.timedelta(days=30 * 6)):
"""Add video logs for the given topics, for each of the given users.
If no users are given, they are created.
If no topics exist, they are taken from the list at the top of this file."""
own_device = Device.get_own_device()
date_diff = datetime.datetime.now() - start_date
video_logs = []
# It's not a user: probably a list.
# Recursive case
if not hasattr(facility_user, "username"):
# It's NONE :-/ generate the users first!
if not facility_user:
(facility_user, _, _) = generate_fake_facility_users()
for topic in topics:
for user in facility_user:
video_logs.append(
generate_fake_video_logs(facility_user=user,
topics=[topic],
start_date=start_date))
# Actually generate!
else:
# First, make videos for the associated logs
# Then make some unassociated videos, to simulate both exploration
# and watching videos without finishing.
# Get (or create) user type
try:
user_settings = json.loads(facility_user.notes)
except:
user_settings = sample_user_settings()
facility_user.notes = json.dumps(user_settings)
try:
facility_user.save()
except Exception as e:
logging.error("Error saving facility user: %s" % e)
date_diff_started = datetime.timedelta(
seconds=datediff(date_diff, units="seconds") *
user_settings["time_in_program"]
) # when this user started in the program, relative to NOW
for topic in topics:
videos = get_topic_videos(topic_id=topic)
exercises = get_topic_exercises(topic_id=topic)
exercise_ids = [
ex["id"] if "id" in ex else ex['name'] for ex in exercises
]
exercise_logs = ExerciseLog.objects.filter(user=facility_user,
id__in=exercise_ids)
# Probability of watching a video, irrespective of the context
p_video_outer = probability_of("video",
user_settings=user_settings)
logging.debug(
"# videos: %d; p(videos)=%4.3f, user settings: %s\n" %
(len(videos), p_video_outer, json.dumps(user_settings)))
for video in videos:
p_completed = probability_of("completed",
user_settings=user_settings)
# If we're just doing random videos, fine.
# If these videos relate to exercises, then suppress non-exercise-related videos
# for this user.
p_video = p_video_outer # start with the context-free value
did_exercise = False
if exercise_logs.count() > 0:
# 5x less likely to watch a video if you haven't done the exercise,
if "related_exercise" not in video:
p_video /= 5 # suppress
# 5x more likely to watch a video if they've done the exercise
# 2x more likely to have finished it.
else:
exercise_log = ExerciseLog.objects.filter(
user=facility_user,
id=video["related_exercise"]["id"])
did_exercise = exercise_log.count() != 0
if did_exercise:
p_video *= 5
p_completed *= 2
# Do the sampling
if p_video < random.random():
continue
# didn't watch it
elif p_completed > random.random():
pct_completed = 100.
else: # Slower students will use videos more. Effort also important.
pct_completed = 100. * min(
1.,
sqrt(random.random() *
sqrt(user_settings["effort_level"] *
user_settings["time_in_program"] /
sqrt(user_settings["speed_of_learning"]))))
# Compute quantities based on sample
total_seconds_watched = int(video["duration"] * pct_completed /
100.)
points = int(750 * pct_completed / 100.)
# Choose a rate of videos, based on their effort level.
# Compute the latest possible start time.
# Then sample a start time between their start time
# and the latest possible start_time
if did_exercise:
# More jitter if you learn fast, less jitter if you try harder (more diligent)
date_jitter = datetime.timedelta(days=max(
0,
random.gauss(
1, user_settings["speed_of_learning"] /
user_settings["effort_level"])))
date_completed = exercise_log[
0].completion_timestamp - date_jitter
else:
rate_of_videos = 0.66 * user_settings[
"effort_level"] + 0.33 * user_settings[
"speed_of_learning"] # exercises per day
time_for_watching = total_seconds_watched
time_delta_completed = datetime.timedelta(
seconds=random.randint(
int(time_for_watching),
int(datediff(date_diff_started, units="seconds"))))
date_completed = datetime.datetime.now(
) - time_delta_completed
try:
vlog = VideoLog.objects.get(user=facility_user,
video_id=video["id"])
except VideoLog.DoesNotExist:
logging.info(
"Creating video log: %-12s: %-45s (%4.1f%% watched, %d points)%s"
% (
facility_user.first_name,
video["title"],
pct_completed,
points,
" COMPLETE on %s!" %
date_completed if pct_completed == 100 else "",
))
vlog = VideoLog(
user=facility_user,
video_id=video["id"],
youtube_id=video["youtube_id"],
total_seconds_watched=total_seconds_watched,
points=points,
complete=(pct_completed == 100.),
completion_timestamp=date_completed,
)
try:
vlog.save(update_userlog=False) # avoid userlog issues
except Exception as e:
logging.error("Error saving video log: %s" % e)
continue
video_logs.append(vlog)
return video_logs
def generate_fake_exercise_logs(facility_user=None,
topics=topics,
start_date=datetime.datetime.now() -
datetime.timedelta(days=30 * 6)):
"""Add exercise logs for the given topics, for each of the given users.
If no users are given, they are created.
If no topics exist, they are taken from the list at the top of this file.
By default, users start learning randomly between 6 months ago and now.
"""
date_diff = datetime.datetime.now() - start_date
exercise_logs = []
user_logs = []
# It's not a user: probably a list.
# Recursive case
if not hasattr(facility_user, "username"):
# It's NONE :-/ generate the users first!
if not facility_user:
(facility_user, _, _) = generate_fake_facility_users()
for topic in topics:
for user in facility_user:
(elogs,
ulogs) = generate_fake_exercise_logs(facility_user=user,
topics=[topic],
start_date=start_date)
exercise_logs.append(elogs)
user_logs.append(ulogs)
# Actually generate!
else:
# Get (or create) user type
try:
user_settings = json.loads(facility_user.notes)
except:
user_settings = sample_user_settings()
facility_user.notes = json.dumps(user_settings)
facility_user.save()
date_diff_started = datetime.timedelta(
seconds=datediff(date_diff, units="seconds") *
user_settings["time_in_program"]
) # when this user started in the program, relative to NOW
for topic in topics:
# Get all exercises related to the topic
exercises = get_topic_exercises(topic_id=topic)
# Problem:
# Not realistic for students to have lots of unfinished exercises.
# If they start them, they tend to get stuck, right?
#
# So, need to make it more probable that they will finish an exercise,
# and less probable that they start one.
#
# What we need is P(streak|started), not P(streak)
# Probability of doing any particular exercise
p_exercise = probability_of(qty="exercise",
user_settings=user_settings)
logging.info(
"# exercises: %d; p(exercise)=%4.3f, user settings: %s\n" %
(len(exercises), p_exercise, json.dumps(user_settings)))
# of exercises is related to
for j, exercise in enumerate(exercises):
if random.random() > p_exercise:
continue
# Probability of completing this exercise, and .. proportion of attempts
p_attempts = probability_of(qty="attempts",
user_settings=user_settings)
attempts = int(random.random() * p_attempts * 30 +
10) # always enough to have completed
elog, created = ExerciseLog.objects.get_or_create(
user=facility_user, exercise_id=exercise["id"])
alogs = []
for i in range(0, attempts):
alog = AttemptLog.objects.create(
user=facility_user,
exercise_id=exercise["id"],
timestamp=start_date + date_diff * i / attempts)
alogs.append(alog)
if random.random() < user_settings["speed_of_learning"]:
alog.correct = True
alog.points = 10
alog.save()
elog.attempts = attempts
elog.latest_activity_timestamp = start_date + date_diff
elog.streak_progress = sum([log.correct
for log in alogs][-10:]) * 10
elog.points = sum([log.points for log in alogs][-10:])
elog.save()
exercise_logs.append(elog)
ulog = UserLog(
user=facility_user,
activity_type=1,
start_datetime=start_date,
end_datetime=start_date + date_diff,
last_active_datetime=start_date + date_diff,
)
ulog.save()
user_logs.append(ulog)
return (exercise_logs, user_logs)
def generate_fake_exercise_logs(facility_user=None,
topics=topics,
start_date=datetime.datetime.now() -
datetime.timedelta(days=30 * 6)):
"""Add exercise logs for the given topics, for each of the given users.
If no users are given, they are created.
If no topics exist, they are taken from the list at the top of this file.
By default, users start learning randomly between 6 months ago and now.
"""
own_device = Device.get_own_device()
date_diff = datetime.datetime.now() - start_date
exercise_logs = []
user_logs = []
# It's not a user: probably a list.
# Recursive case
if not hasattr(facility_user, "username"):
# It's NONE :-/ generate the users first!
if not facility_user:
(facility_user, _, _) = generate_fake_facility_users()
for topic in topics:
for user in facility_user:
(elogs,
ulogs) = generate_fake_exercise_logs(facility_user=user,
topics=[topic],
start_date=start_date)
exercise_logs.append(elogs)
user_logs.append(ulogs)
# Actually generate!
else:
# Get (or create) user type
try:
user_settings = json.loads(facility_user.notes)
except:
user_settings = sample_user_settings()
facility_user.notes = json.dumps(user_settings)
facility_user.save()
date_diff_started = datetime.timedelta(
seconds=datediff(date_diff, units="seconds") *
user_settings["time_in_program"]
) # when this user started in the program, relative to NOW
for topic in topics:
# Get all exercises related to the topic
exercises = get_topic_exercises(topic_id=topic)
# Problem:
# Not realistic for students to have lots of unfinished exercises.
# If they start them, they tend to get stuck, right?
#
# So, need to make it more probable that they will finish an exercise,
# and less probable that they start one.
#
# What we need is P(streak|started), not P(streak)
# Probability of doing any particular exercise
p_exercise = probability_of(qty="exercise",
user_settings=user_settings)
logging.debug(
"# exercises: %d; p(exercise)=%4.3f, user settings: %s\n" %
(len(exercises), p_exercise, json.dumps(user_settings)))
# of exercises is related to
for j, exercise in enumerate(exercises):
if random.random() > p_exercise:
continue
# Probability of completing this exercise, and .. proportion of attempts
p_completed = probability_of(qty="completed",
user_settings=user_settings)
p_attempts = probability_of(qty="attempts",
user_settings=user_settings)
attempts = int(random.random() * p_attempts * 30 +
10) # always enough to have completed
completed = (random.random() < p_completed)
if completed:
streak_progress = 100
else:
streak_progress = max(
0,
min(
90,
random.gauss(
100 * user_settings["speed_of_learning"], 20)))
streak_progress = int(floor(streak_progress / 10.)) * 10
points = streak_progress / 10 * 12 if completed else 0 # only get points when you master.
# Choose a rate of exercises, based on their effort level and speed of learning.
# Compute the latest possible start time.
# Then sample a start time between their start time
# and the latest possible start_time
rate_of_exercises = 0.66 * user_settings[
"effort_level"] + 0.33 * user_settings[
"speed_of_learning"] # exercises per day
time_for_attempts = min(
datetime.timedelta(days=rate_of_exercises * attempts),
date_diff_started) # protect with min
time_delta_completed = datetime.timedelta(
seconds=random.randint(
int(datediff(time_for_attempts, units="seconds")),
int(datediff(date_diff_started, units="seconds"))))
date_completed = datetime.datetime.now() - time_delta_completed
# Always create new
logging.info(
"Creating exercise log: %-12s: %-25s (%d points, %d attempts, %d%% streak on %s)"
% (
facility_user.first_name,
exercise["name"],
points,
attempts,
streak_progress,
date_completed,
))
try:
elog = ExerciseLog.objects.get(
user=facility_user, exercise_id=exercise["name"])
except ExerciseLog.DoesNotExist:
elog = ExerciseLog(
user=facility_user,
exercise_id=exercise["name"],
attempts=int(attempts),
streak_progress=streak_progress,
points=int(points),
complete=completed,
completion_timestamp=date_completed,
)
try:
elog.save(update_userlog=False)
# For now, make all attempts on an exercise into a single UserLog.
seconds_per_attempt = 10 * (
1 + user_settings["speed_of_learning"] *
random.random())
time_to_navigate = 15 * (0.5 + random.random()
) #between 7.5s and 22.5s
time_to_logout = 5 * (0.5 + random.random()
) # between 2.5 and 7.5s
if UserLog.is_enabled():
ulog = UserLog(
user=facility_user,
activity_type=1,
start_datetime=date_completed -
datetime.timedelta(seconds=int(
attempts * seconds_per_attempt +
time_to_navigate)),
end_datetime=date_completed +
datetime.timedelta(seconds=time_to_logout),
last_active_datetime=date_completed,
)
ulog.save()
user_logs.append(ulog)
except Exception as e:
logging.error("Error saving exercise log: %s" % e)
continue
exercise_logs.append(elog)
return (exercise_logs, user_logs)
def generate_fake_exercise_logs(
facility_user=None, topics=topics, start_date=datetime.datetime.now() - datetime.timedelta(days=30 * 6)
):
"""Add exercise logs for the given topics, for each of the given users.
If no users are given, they are created.
If no topics exist, they are taken from the list at the top of this file.
By default, users start learning randomly between 6 months ago and now.
"""
date_diff = datetime.datetime.now() - start_date
exercise_logs = []
user_logs = []
# It's not a user: probably a list.
# Recursive case
if not hasattr(facility_user, "username"):
# It's NONE :-/ generate the users first!
if not facility_user:
(facility_user, _, _) = generate_fake_facility_users()
for topic in topics:
for user in facility_user:
(elogs, ulogs) = generate_fake_exercise_logs(facility_user=user, topics=[topic], start_date=start_date)
exercise_logs.append(elogs)
user_logs.append(ulogs)
# Actually generate!
else:
# Get (or create) user type
try:
user_settings = json.loads(facility_user.notes)
except:
user_settings = sample_user_settings()
facility_user.notes = json.dumps(user_settings)
facility_user.save()
date_diff_started = datetime.timedelta(
seconds=datediff(date_diff, units="seconds") * user_settings["time_in_program"]
) # when this user started in the program, relative to NOW
for topic in topics:
# Get all exercises related to the topic
exercises = get_topic_exercises(topic_id=topic)
# Problem:
# Not realistic for students to have lots of unfinished exercises.
# If they start them, they tend to get stuck, right?
#
# So, need to make it more probable that they will finish an exercise,
# and less probable that they start one.
#
# What we need is P(streak|started), not P(streak)
# Probability of doing any particular exercise
p_exercise = probability_of(qty="exercise", user_settings=user_settings)
logging.info(
"# exercises: %d; p(exercise)=%4.3f, user settings: %s\n"
% (len(exercises), p_exercise, json.dumps(user_settings))
)
# of exercises is related to
for j, exercise in enumerate(exercises):
if random.random() > p_exercise:
continue
# Probability of completing this exercise, and .. proportion of attempts
p_attempts = probability_of(qty="attempts", user_settings=user_settings)
attempts = int(random.random() * p_attempts * 30 + 10) # always enough to have completed
elog, created = ExerciseLog.objects.get_or_create(user=facility_user, exercise_id=exercise["id"])
alogs = []
for i in range(0, attempts):
alog = AttemptLog.objects.create(
user=facility_user, exercise_id=exercise["id"], timestamp=start_date + date_diff * i / attempts
)
alogs.append(alog)
if random.random() < user_settings["speed_of_learning"]:
alog.correct = True
alog.points = 10
alog.save()
elog.attempts = attempts
elog.latest_activity_timestamp = start_date + date_diff
elog.streak_progress = sum([log.correct for log in alogs][-10:]) * 10
elog.points = sum([log.points for log in alogs][-10:])
elog.save()
exercise_logs.append(elog)
ulog = UserLog(
user=facility_user,
activity_type=1,
start_datetime=start_date,
end_datetime=start_date + date_diff,
last_active_datetime=start_date + date_diff,
)
ulog.save()
user_logs.append(ulog)
return (exercise_logs, user_logs)
def generate_fake_exercise_logs(facility_user=None, topics=topics, start_date=datetime.datetime.now() - datetime.timedelta(days=30 * 6)):
"""Add exercise logs for the given topics, for each of the given users.
If no users are given, they are created.
If no topics exist, they are taken from the list at the top of this file.
By default, users start learning randomly between 6 months ago and now.
"""
date_diff = datetime.datetime.now() - start_date
exercise_logs = []
user_logs = []
# It's not a user: probably a list.
# Recursive case
if not hasattr(facility_user, "username"):
# It's NONE :-/ generate the users first!
if not facility_user:
(facility_user, _, _) = generate_fake_facility_users()
for topic in topics:
for user in facility_user:
(elogs, ulogs) = generate_fake_exercise_logs(facility_user=user, topics=[topic], start_date=start_date)
exercise_logs.append(elogs)
user_logs.append(ulogs)
# Actually generate!
else:
# Get (or create) user type
try:
user_settings = json.loads(facility_user.notes)
except:
user_settings = sample_user_settings()
facility_user.notes = json.dumps(user_settings)
facility_user.save()
date_diff_started = datetime.timedelta(seconds=datediff(date_diff, units="seconds") * user_settings["time_in_program"]) # when this user started in the program, relative to NOW
for topic in topics:
# Get all exercises related to the topic
exercises = get_topic_exercises(topic_id=topic)
# Problem:
# Not realistic for students to have lots of unfinished exercises.
# If they start them, they tend to get stuck, right?
#
# So, need to make it more probable that they will finish an exercise,
# and less probable that they start one.
#
# What we need is P(streak|started), not P(streak)
# Probability of doing any particular exercise
p_exercise = probability_of(qty="exercise", user_settings=user_settings)
logging.debug("# exercises: %d; p(exercise)=%4.3f, user settings: %s\n" % (len(exercises), p_exercise, json.dumps(user_settings)))
# of exercises is related to
for j, exercise in enumerate(exercises):
if random.random() > p_exercise:
continue
# Probability of completing this exercise, and .. proportion of attempts
p_completed = probability_of(qty="completed", user_settings=user_settings)
p_attempts = probability_of(qty="attempts", user_settings=user_settings)
attempts = int(random.random() * p_attempts * 30 + 10) # always enough to have completed
completed = (random.random() < p_completed)
if completed:
streak_progress = 100
else:
streak_progress = max(0, min(90, random.gauss(100 * user_settings["speed_of_learning"], 20)))
streak_progress = int(floor(streak_progress / 10.)) * 10
points = streak_progress / 10 * 12 if completed else 0 # only get points when you master.
# Choose a rate of exercises, based on their effort level and speed of learning.
# Compute the latest possible start time.
# Then sample a start time between their start time
# and the latest possible start_time
rate_of_exercises = 0.66 * user_settings["effort_level"] + 0.33 * user_settings["speed_of_learning"] # exercises per day
time_for_attempts = min(datetime.timedelta(days=rate_of_exercises * attempts), date_diff_started) # protect with min
time_delta_completed = datetime.timedelta(seconds=random.randint(int(datediff(time_for_attempts, units="seconds")), int(datediff(date_diff_started, units="seconds"))))
date_completed = datetime.datetime.now() - time_delta_completed
# Always create new
logging.info("Creating exercise log: %-12s: %-25s (%d points, %d attempts, %d%% streak on %s)" % (
facility_user.first_name,
exercise["name"],
points,
attempts,
streak_progress,
date_completed,
))
try:
elog = ExerciseLog.objects.get(user=facility_user, exercise_id=exercise["name"])
except ExerciseLog.DoesNotExist:
elog = ExerciseLog(
user=facility_user,
exercise_id=exercise["name"],
attempts=int(attempts),
streak_progress=streak_progress,
points=int(points),
complete=completed,
completion_timestamp=date_completed,
)
try:
elog.save()
# For now, make all attempts on an exercise into a single UserLog.
seconds_per_attempt = 10 * (1 + user_settings["speed_of_learning"] * random.random())
time_to_navigate = 15 * (0.5 + random.random()) #between 7.5s and 22.5s
time_to_logout = 5 * (0.5 + random.random()) # between 2.5 and 7.5s
if UserLog.is_enabled():
ulog = UserLog(
user=facility_user,
activity_type=1,
start_datetime = date_completed - datetime.timedelta(seconds=int(attempts * seconds_per_attempt + time_to_navigate)),
end_datetime = date_completed + datetime.timedelta(seconds=time_to_logout),
last_active_datetime = date_completed,
)
ulog.save()
user_logs.append(ulog)
except Exception as e:
logging.error("Error saving exercise log: %s" % e)
continue
exercise_logs.append(elog)
return (exercise_logs, user_logs)
def update_language_packs(lang_codes, options):
package_metadata = {}
since_date = datetime.datetime.now() - datetime.timedelta(int(options["days"]))
if options['update_dubbed']:
# Get the latest dubbed video map; it's shared across language packs
force_dubbed_download = not os.path.exists(DUBBED_VIDEOS_MAPPING_FILEPATH) \
or 0 < datediff(since_date, datetime.datetime.fromtimestamp(os.path.getctime(DUBBED_VIDEOS_MAPPING_FILEPATH)))
get_dubbed_video_map(force=force_dubbed_download)
for lang_code in lang_codes:
lang_code_map = get_supported_language_map(lang_code)
lang_metadata = {
"beta": True if lang_code_map["beta"] != lang_code else False,
}
# Step 1: Update / collect srts. No version needed, we want to share latest always.
if options['update_srts']:
update_srts(since_date=since_date, lang_codes=[lang_code_map["amara"]])
lang_metadata["subtitle_count"] = get_subtitle_count(lang_code_map["amara"])
# Step 2: Update the dubbed video mappings. No version needed, we want to share latest always.
dv_map = get_dubbed_video_map(lang_code_map["dubbed_videos"])
lang_metadata["num_dubbed_videos"] = len(dv_map) if dv_map and version_diff(options["version"], COMMAND_MIN_VERSION) > 0 else 0
# Step 3: Update the exercises. No version needed, we want to share latest always.
# TODO(bcipolli): make sure that each language pack only grabs exercises that are included in its topic tree.
if options['update_exercises'] and version_diff(options["version"], COMMAND_MIN_VERSION) > 0:
call_command("scrape_exercises", lang_code=lang_code_map["exercises"])
lang_metadata["num_exercises"] = get_localized_exercise_count(lang_code_map["exercises"]) if version_diff(options["version"], COMMAND_MIN_VERSION) > 0 else 0
# Step 4: Update the crowdin translations. Version needed!
# TODO(bcipolli): skip this when we're going backwards in version.
if options["no_update"] or version_diff(options["version"], COMMAND_MIN_VERSION) == 0:
trans_metadata = {lang_code: get_po_metadata(get_po_build_path(lang_code))}
else:
try:
trans_metadata = update_translations(
lang_codes=[lang_code], # will be converted, as needed
zip_file=options['zip_file'],
ka_zip_file=options['ka_zip_file'],
download_ka_translations=options['update_ka_trans'],
download_kalite_translations=options['update_kalite_trans'],
use_local=options["use_local"],
version=options["version"],
)
except SkipTranslations:
trans_metadata = {lang_code: get_po_metadata(get_po_build_path(lang_code))}
lang_metadata.update(trans_metadata.get(lang_code, {}))
# Now create/update unified meta data
generate_metadata(package_metadata={lang_code: lang_metadata}, version=options["version"], force_version_update=options["force_update"])
# Zip into language packs
package_sizes = zip_language_packs(lang_codes=[lang_code], version=options["version"])
logging.debug("%s sizes: %s" % (lang_code, package_sizes.get(lang_code, {})))
lang_metadata.update(package_sizes.get(lang_code, {}))
# Update the metadata with the package size information
update_metadata({lang_code: lang_metadata}, version=options["version"])
# Update package metadata
package_metadata[lang_code] = lang_metadata
return package_metadata
def update_language_packs(lang_codes, options):
package_metadata = {}
since_date = datetime.datetime.now() - datetime.timedelta(int(options["days"]))
if options['update_dubbed']:
# Get the latest dubbed video map; it's shared across language packs
force_dubbed_download = not os.path.exists(DUBBED_VIDEOS_MAPPING_FILEPATH) \
or 0 < datediff(since_date, datetime.datetime.fromtimestamp(os.path.getctime(DUBBED_VIDEOS_MAPPING_FILEPATH)))
get_dubbed_video_map(force=force_dubbed_download)
for lang_code in lang_codes:
lang_code_map = get_supported_language_map(lang_code)
lang_metadata = {}
# Step 1: Update / collect srts. No version needed, we want to share latest always.
if options['update_srts']:
update_srts(since_date=since_date, lang_codes=[lang_code_map["amara"]])
lang_metadata["subtitle_count"] = get_subtitle_count(lang_code_map["amara"])
# Step 2: Update the dubbed video mappings. No version needed, we want to share latest always.
dv_map = get_dubbed_video_map(lang_code_map["dubbed_videos"])
lang_metadata["num_dubbed_videos"] = len(dv_map) if dv_map and version_diff(options["version"], "0.10.3") > 0 else 0
# Step 3: Update the exercises. No version needed, we want to share latest always.
# TODO(bcipolli): make sure that each language pack only grabs exercises that are included in its topic tree.
if options['update_exercises'] and version_diff(options["version"], "0.10.3") > 0:
call_command("scrape_exercises", lang_code=lang_code_map["exercises"])
lang_metadata["num_exercises"] = get_localized_exercise_count(lang_code_map["exercises"]) if version_diff(options["version"], "0.10.3") > 0 else 0
# Step 4: Update the crowdin translations. Version needed!
# TODO(bcipolli): skip this when we're going backwards in version.
if options["no_update"] or version_diff(options["version"], "0.10.3") == 0:
trans_metadata = {lang_code: get_po_metadata(get_po_build_path(lang_code))}
else:
try:
trans_metadata = update_translations(
lang_codes=[lang_code], # will be converted, as needed
zip_file=options['zip_file'],
ka_zip_file=options['ka_zip_file'],
download_ka_translations=options['update_ka_trans'],
download_kalite_translations=options['update_kalite_trans'],
use_local=options["use_local"],
version=options["version"],
)
except SkipTranslations:
trans_metadata = {lang_code: get_po_metadata(get_po_build_path(lang_code))}
lang_metadata.update(trans_metadata.get(lang_code, {}))
# Now create/update unified meta data
generate_metadata(package_metadata={lang_code: lang_metadata}, version=options["version"], force_version_update=options["force_update"])
# Zip into language packs
package_sizes = zip_language_packs(lang_codes=[lang_code], version=options["version"])
logging.debug("%s sizes: %s" % (lang_code, package_sizes.get(lang_code, {})))
lang_metadata.update(package_sizes.get(lang_code, {}))
# Update the metadata with the package size information
update_metadata({lang_code: lang_metadata}, version=options["version"])
# Update package metadata
package_metadata[lang_code] = lang_metadata
return package_metadata
def generate_fake_video_logs(
facility_user=None, topics=topics, start_date=datetime.datetime.now() - datetime.timedelta(days=30 * 6)
):
"""Add video logs for the given topics, for each of the given users.
If no users are given, they are created.
If no topics exist, they are taken from the list at the top of this file."""
date_diff = datetime.datetime.now() - start_date
video_logs = []
# It's not a user: probably a list.
# Recursive case
if not hasattr(facility_user, "username"):
# It's NONE :-/ generate the users first!
if not facility_user:
(facility_user, _, _) = generate_fake_facility_users()
for topic in topics:
for user in facility_user:
video_logs.append(generate_fake_video_logs(facility_user=user, topics=[topic], start_date=start_date))
# Actually generate!
else:
# First, make videos for the associated logs
# Then make some unassociated videos, to simulate both exploration
# and watching videos without finishing.
# Get (or create) user type
try:
user_settings = json.loads(facility_user.notes)
except:
user_settings = sample_user_settings()
facility_user.notes = json.dumps(user_settings)
try:
facility_user.save()
except Exception as e:
logging.error("Error saving facility user: %s" % e)
date_diff_started = datetime.timedelta(
seconds=datediff(date_diff, units="seconds") * user_settings["time_in_program"]
) # when this user started in the program, relative to NOW
# contains the video duration key
video_cache = get_content_cache()
for topic in topics:
videos = get_topic_videos(topic_id=topic)
exercises = get_topic_exercises(topic_id=topic)
exercise_ids = [ex["id"] if "id" in ex else ex["name"] for ex in exercises]
exercise_logs = ExerciseLog.objects.filter(user=facility_user, id__in=exercise_ids)
# Probability of watching a video, irrespective of the context
p_video_outer = probability_of("video", user_settings=user_settings)
logging.debug(
"# videos: %d; p(videos)=%4.3f, user settings: %s\n"
% (len(videos), p_video_outer, json.dumps(user_settings))
)
for video in videos:
p_completed = probability_of("completed", user_settings=user_settings)
# If we're just doing random videos, fine.
# If these videos relate to exercises, then suppress non-exercise-related videos
# for this user.
p_video = p_video_outer # start with the context-free value
did_exercise = False
if exercise_logs.count() > 0:
# 5x less likely to watch a video if you haven't done the exercise,
if "related_exercise" not in video:
p_video /= 5 # suppress
# 5x more likely to watch a video if they've done the exercise
# 2x more likely to have finished it.
else:
exercise_log = ExerciseLog.objects.filter(
user=facility_user, id=video["related_exercise"]["id"]
)
did_exercise = exercise_log.count() != 0
if did_exercise:
p_video *= 5
p_completed *= 2
# Do the sampling
if p_video < random.random():
continue
# didn't watch it
elif p_completed > random.random():
pct_completed = 100.0
else: # Slower students will use videos more. Effort also important.
pct_completed = 100.0 * min(
1.0,
sqrt(
random.random()
* sqrt(
user_settings["effort_level"]
* user_settings["time_in_program"]
/ sqrt(user_settings["speed_of_learning"])
)
),
)
# get the video duration on the video cache
video_id = video.get("id", "")
video_duration = 0
if video_id and video_cache:
video_item = video_cache.get(video_id, None)
if video_item:
video_duration = video_item.get("duration", 0)
# Compute quantities based on sample
total_seconds_watched = int(video_duration * pct_completed / 100.0)
points = int(750 * pct_completed / 100.0)
# Choose a rate of videos, based on their effort level.
# Compute the latest possible start time.
# Then sample a start time between their start time
# and the latest possible start_time
if did_exercise:
# More jitter if you learn fast, less jitter if you try harder (more diligent)
date_jitter = datetime.timedelta(
days=max(0, random.gauss(1, user_settings["speed_of_learning"] / user_settings["effort_level"]))
)
date_completed = exercise_log[0].completion_timestamp - date_jitter
else:
rate_of_videos = (
0.66 * user_settings["effort_level"] + 0.33 * user_settings["speed_of_learning"]
) # exercises per day
time_for_watching = total_seconds_watched
time_delta_completed = datetime.timedelta(
seconds=random.randint(
int(time_for_watching), int(datediff(date_diff_started, units="seconds"))
)
)
date_completed = datetime.datetime.now() - time_delta_completed
try:
vlog = VideoLog.objects.get(user=facility_user, video_id=video_id)
except VideoLog.DoesNotExist:
logging.info(
"Creating video log: %-12s: %-45s (%4.1f%% watched, %d points)%s"
% (
facility_user.first_name,
video["title"],
pct_completed,
points,
" COMPLETE on %s!" % date_completed if pct_completed == 100 else "",
)
)
youtube_id = video.get("youtube_id", video_id)
vlog = VideoLog(
user=facility_user,
video_id=video_id,
youtube_id=youtube_id,
total_seconds_watched=total_seconds_watched,
points=points,
complete=(pct_completed == 100.0),
completion_timestamp=date_completed,
latest_activity_timestamp=date_completed,
)
try:
vlog.save() # avoid userlog issues
except Exception as e:
logging.error("Error saving video log: %s" % e)
continue
video_logs.append(vlog)
return video_logs
def test_units(self):
"""
A single difference, tested across different units
"""
d1 = datetime.datetime(2000, 1, 1)
d2 = datetime.datetime(2004, 1, 1) # test 4 years apart
self.assertEqual(datediff(d2, d1, units="microsecond"), 1E6*60*60*24*(365*4 + 1), "4 years (with leap year), in microseconds")
self.assertEqual(datediff(d2, d1, units="microseconds"), 1E6*60*60*24*(365*4 + 1), "4 years (with leap year), in microseconds")
self.assertEqual(datediff(d2, d1), 60*60*24*(365*4 + 1), "4 years (with leap year), in seconds (default)")
self.assertEqual(datediff(d2, d1, units="second"), 60*60*24*(365*4 + 1), "4 years (with leap year), in seconds")
self.assertEqual(datediff(d2, d1, units="seconds"), 60*60*24*(365*4 + 1), "4 years (with leap year), in seconds")
self.assertEqual(datediff(d2, d1, units="minute"), 60*24*(365*4 + 1), "4 years (with leap year), in minutes")
self.assertEqual(datediff(d2, d1, units="minutes"), 60*24*(365*4 + 1), "4 years (with leap year), in minutes")
self.assertEqual(datediff(d2, d1, units="hour"), 24*(365*4 + 1), "4 years (with leap year), in hours")
self.assertEqual(datediff(d2, d1, units="hours"), 24*(365*4 + 1), "4 years (with leap year), in hours")
self.assertEqual(datediff(d2, d1, units="day"), 365*4 + 1, "4 years (with leap year), in days")
self.assertEqual(datediff(d2, d1, units="days"), 365*4 + 1, "4 years (with leap year), in days")
self.assertEqual(datediff(d2, d1, units="week"), (365*4 + 1)/7., "4 years (with leap year), in weeks")
self.assertEqual(datediff(d2, d1, units="weeks"), (365*4 + 1)/7., "4 years (with leap year), in weeks")