def get_probability():
msg = request.args.get('message')
if not title:
return json.dumps({'error': 'no \'message\' argument given'})
prob = util.probability(msg)
response = {'message': msg, 'p': prob}
return json.dumps(response)
def get_probability():
msg = request.args.get("message")
if not title:
return json.dumps({"error": "no 'message' argument given"})
prob = util.probability(msg)
response = {"message": msg, "p": prob}
return json.dumps(response)
def main():
try:
exercise_file = open(util.relative_path("exercise_reports"), 'r+')
ex_reports = json.loads(exercise_file.read())
except IOError:
exercise_file = open(util.relative_path("exercise_reports"), 'w')
ex_reports = {"elapsed_time": 1, # Filler value
"max_id": -1,
"last_time": 0}
new_reports = get_errors(copy.deepcopy(ex_reports))
period_len = new_reports["time_this_period"]
for ex in new_reports:
if ex in SPECIAL_VALUES:
continue
if ex in ex_reports and ex_reports[ex]["num_errors"] > 0:
errors_this_period = new_reports[ex]["this_period"]
mean, probability = util.probability(ex_reports[ex]["num_errors"],
ex_reports["elapsed_time"],
errors_this_period,
period_len)
print ("%s] TOTAL %s/%ss; %s-: %s/%ss; m=%.3f p=%.3f"
% (time.strftime("%Y-%m-%d %H:%M:%S %Z"),
ex_reports[ex]["num_errors"], ex_reports["elapsed_time"],
ex_reports["last_time"],
errors_this_period, period_len,
mean, probability))
if (probability > 0.997 and errors_this_period > 1):
util.send_to_slack(
"*Elevated exercise bug report rate in exercise `%s`\n"
"Reports: %s. We saw %s in the last %s minutes,"
" while the mean indicates we should see around %s."
" *Probability that this is abnormally elevated: %.4f.*"
% (ex,
generate_slack_links(new_reports[ex]["href"]),
util.thousand_commas(errors_this_period),
util.thousand_commas(int(period_len / 60)),
util.thousand_commas(round(mean, 2)),
probability),
channel="#support")
if "href" in new_reports[ex].keys():
del new_reports[ex]["href"] # don't need to keep the links around
del new_reports["time_this_period"]
# Overwrite with new contents
exercise_file.seek(0)
exercise_file.truncate()
exercise_file.write(json.dumps(new_reports))
exercise_file.close()
def main():
try:
exercise_file = open(util.relative_path("exercise_reports"), 'r+')
ex_reports = json.loads(exercise_file.read())
except IOError:
exercise_file = open(util.relative_path("exercise_reports"), 'w')
ex_reports = {
"elapsed_time": 1, # Filler value
"max_id": -1,
"last_time": 0
}
new_reports = get_errors(copy.deepcopy(ex_reports))
period_len = new_reports["time_this_period"]
for ex in new_reports:
if ex in SPECIAL_VALUES:
continue
if ex in ex_reports and ex_reports[ex]["num_errors"] > 0:
errors_this_period = new_reports[ex]["this_period"]
mean, probability = util.probability(ex_reports[ex]["num_errors"],
ex_reports["elapsed_time"],
errors_this_period,
period_len)
print("%s] TOTAL %s/%ss; %s-: %s/%ss; m=%.3f p=%.3f" %
(time.strftime("%Y-%m-%d %H:%M:%S %Z"),
ex_reports[ex]["num_errors"], ex_reports["elapsed_time"],
ex_reports["last_time"], errors_this_period, period_len,
mean, probability))
if (probability > 0.997 and errors_this_period > 1):
util.send_to_slack(
"*Elevated exercise bug report rate in exercise `%s`\n"
"Reports: %s. We saw %s in the last %s minutes,"
" while the mean indicates we should see around %s."
" *Probability that this is abnormally elevated: %.4f.*" %
(ex, generate_slack_links(new_reports[ex]["href"]),
util.thousand_commas(errors_this_period),
util.thousand_commas(int(period_len / 60)),
util.thousand_commas(round(mean, 2)), probability),
channel="#support")
if "href" in new_reports[ex].keys():
del new_reports[ex]["href"] # don't need to keep the links around
del new_reports["time_this_period"]
# Overwrite with new contents
exercise_file.seek(0)
exercise_file.truncate()
exercise_file.write(json.dumps(new_reports))
exercise_file.close()
def main():
try:
exercise_file = open(util.relative_path("exercise_reports"), 'r+')
ex_reports = json.loads(exercise_file.read())
except IOError:
exercise_file = open(util.relative_path("exercise_reports"), 'w')
ex_reports = {"elapsed_time": 1, # Filler value
"max_id": -1,
"last_time": 0}
new_reports = get_errors(copy.deepcopy(ex_reports))
period_len = new_reports["time_this_period"]
for ex in new_reports:
if ex in SPECIAL_VALUES:
continue
if ex in ex_reports and ex_reports[ex]["num_errors"] > 0:
errors_this_period = new_reports[ex]["this_period"]
mean, probability = util.probability(ex_reports[ex]["num_errors"],
ex_reports["elapsed_time"],
errors_this_period,
period_len)
if (probability > 0.997 and errors_this_period > 1):
# Too many errors!
hipchat_message.send_message(
"Elevated exercise bug report rate in exercise %s!"
" Reports: %s. We saw %s in the last %s minutes,"
" while the mean indicates we should see around %s."
" Probability that this is abnormally elevated: %.4f."
% (ex,
generate_links(new_reports[ex]["href"]),
util.thousand_commas(errors_this_period),
util.thousand_commas(int(period_len / 60)),
util.thousand_commas(round(mean, 2)),
probability),
room_id="Exercises")
if "href" in new_reports[ex].keys():
del new_reports[ex]["href"] # don't need to keep the links around
del new_reports["time_this_period"]
# Overwrite with new contents
exercise_file.seek(0)
exercise_file.truncate()
exercise_file.write(json.dumps(new_reports))
exercise_file.close()
def main():
try:
google_code_file = open(util.relative_path("google_code"), 'r+')
old_reports = json.loads(google_code_file.read())
except IOError:
google_code_file = open(util.relative_path("google_code"), 'w')
# elapsed_time is filler value: doesn't matter what it is
# since issue_count is 0.
old_reports = {"elapsed_time": 1,
"last_id": -1,
"issue_count": 0,
"last_time": 0}
new_reports = get_errors(copy.deepcopy(old_reports))
time_this_period = new_reports["time_this_period"]
mean, probability = util.probability(old_reports["issue_count"],
old_reports["elapsed_time"],
new_reports["issues_this_period"],
time_this_period)
if (mean != 0 and probability > 0.99):
# Too many errors!
hipchat_message.send_message(
"Elevated bug report rate on"
" <a href='http://khanacademy.org/r/bugs'>Google"
" code!</a>"
" We saw %s in the last %s minutes,"
" while the mean indicates we should see around %s."
" Probability that this is abnormally elevated: %.4f."
% (util.thousand_commas(new_reports["issues_this_period"]),
util.thousand_commas(int(time_this_period / 60)),
util.thousand_commas(round(mean, 2)),
probability))
# Delete fields we don't need anymore
del(new_reports["issues_this_period"])
del(new_reports["time_this_period"])
google_code_file.seek(0)
google_code_file.truncate()
google_code_file.write(json.dumps(new_reports))
google_code_file.close()
def main():
try:
zendesk_status_file = util.relative_path("zendesk")
with open(zendesk_status_file) as f:
old_data = cPickle.load(f)
except (IOError, EOFError):
old_data = {"elapsed_time_weekday": 0.0001, # avoid a divide-by-0
"elapsed_time_weekend": 0.0001, # avoid a divide-by-0
"ticket_count_weekday": 0,
"ticket_count_weekend": 0,
"last_time_t": None,
"last_time_t_weekday": None,
"last_time_t_weekend": None,
}
# We compare the number of tickets in the last few minutes against
# the historical average for all time. But we don't start "all
# time" at AD 1, we start it a week ago. Longer than that and it
# takes forever due to quota issues. That's still plenty of
# historical data. :-)
#
# Zendesk seems to wait 5 minutes to update API data :-(, so we
# ask for data that's a bit time-lagged
end_time = int(time.time()) - 300
start_time = old_data['last_time_t']
# Set flag to track if current time period is a weekend. Separate
# ticket_count/elapsed_time stats are kept for weekend vs. weekday
# to improve sensitivity to increases during low-traffic periods
is_off_hours = _is_off_hours(datetime.datetime.fromtimestamp(end_time))
(new_tickets, oldest_ticket_time_t) = get_tickets_between(
start_time or (end_time - 86400 * 7), end_time)
num_new_tickets = len(new_tickets)
# The first time we run this, we take the starting time to be the
# time of the first bug report.
if start_time is None:
start_time = oldest_ticket_time_t
time_this_period = end_time - start_time
if is_off_hours:
# To simplify backcompat we still use "weekend" and "weekday" in the
# saved data; really they mean "on hours" and "off hours" now.
ticket_count = old_data['ticket_count_weekend']
elapsed_time = old_data['elapsed_time_weekend']
else:
ticket_count = old_data['ticket_count_weekday']
elapsed_time = old_data['elapsed_time_weekday']
(mean, probability) = util.probability(ticket_count,
elapsed_time,
num_new_tickets,
time_this_period)
print ("%s] TOTAL %s/%ss; %s-: %s/%ss; m=%.3f p=%.3f"
% (time.strftime("%Y-%m-%d %H:%M:%S %Z"),
ticket_count, int(elapsed_time),
start_time,
num_new_tickets, time_this_period,
mean, probability))
handle_alerts(new_tickets, time_this_period, mean, probability,
start_time, end_time)
if is_off_hours:
new_data = {"elapsed_time_weekend": (
old_data["elapsed_time_weekend"] + time_this_period),
"ticket_count_weekend": (
old_data["ticket_count_weekend"] + num_new_tickets),
"elapsed_time_weekday": old_data["elapsed_time_weekday"],
"ticket_count_weekday": old_data["ticket_count_weekday"],
}
else:
new_data = {"elapsed_time_weekend": old_data["elapsed_time_weekend"],
"ticket_count_weekend": old_data["ticket_count_weekend"],
"elapsed_time_weekday": (
old_data["elapsed_time_weekday"] + time_this_period),
"ticket_count_weekday": (
old_data["ticket_count_weekday"] + num_new_tickets),
}
new_data['last_time_t'] = end_time
with open(zendesk_status_file, 'w') as f:
cPickle.dump(new_data, f)
THRESH = 0.05
# Repeat indefinitely
while 1:
# Capture the input from stdin, which is the stream of data from ingest.py.
# It is dumped as JSON, so decode it.
line = stdin.readline()
edit = json.loads(line)
# We are extracting the 'wiki' key, which is a unique identifier for the
# Wikipedia that was edited.
message = edit.get('wiki')
# I have written a function in util that gets the probability of a
# particular message, given the entries in the Redis database.
prob = util.probability(message)
# If the probability falls below our threshold, emit a message. Otherwise,
# loop around.
if prob < THRESH:
# This schema (particularly the 'unlikely_message' type) is understood
# by the slack.py file, which sends the appropriate alerts.
alert = {
'type': 'unlikely_message',
'message': message,
'prob': prob
}
# Print the alert to stdout and flush stdout to prevent message delay
# from buffering.
print(json.dumps(alert))
stdout.flush()
n_PC) + 'PC/control/'
name_out = mode + '_control_' + 'traj_' + str(traj_n)
plt.savefig(path_out + name_out + ".png")
plt.show()
########## Retrieve maxiumum number of elements based on a raw decimal discretization ##########
data = mPE_vector.flatten()
kmeans = KMeans(n_clusters=20).fit(data.reshape(-1, 1))
kmeans.predict(data.reshape(-1, 1))
centroids = kmeans.cluster_centers_
centroids = [centroids[i] for i in range(len(centroids))]
centroids = np.asarray(centroids)
########## Get probability vectors ##########
prob1 = probability(mPE_vector[0, :, 0], centroids)
prob2 = probability(mPE_vector[1, :, 0], centroids)
########## Evaluation ##########
significance_lev = significance_level
js_distance = distance.jensenshannon(prob1, prob2)
[_, p_value] = stats.ks_2samp(mPE_vector[0, :, 0], mPE_vector[1, :,
0])
if p_value > 0.8:
too_low = False
break
measures[1, 0] = p_value
measures[2, 0] = js_distance[0]
def main():
try:
jira_status_file = util.relative_path('jira')
with open(jira_status_file) as f:
old_data = cPickle.load(f)
except IOError:
old_data = {'elapsed_times': {},
'ticket_counts': collections.defaultdict(int),
'last_time_t': None,
}
# We compare the number of tickets in the last few minutes against
# the historical average for all time. But we don't start "all
# time" at AD 1, we start it 100 days ago.
# Note: this is a way wider window than we use for Zendesk, but we're
# making exercise-specific recommendations, so we need more data.
now = int(time.time())
num_days_in_past = 100
(num_new_tickets, oldest_ticket_time_t) = num_tickets_between(
old_data['last_time_t'] or (now - 86400 * num_days_in_past), now)
# Elapsed time is computed per-exercise, so store values as we go.
# We use a copy so that exercises that don't appear as new tickets still
# have their old elapsed times preserved.
elapsed_times = copy.copy(old_data['elapsed_times'])
for exercise in num_new_tickets:
# If this is the first time we're running, we don't have a last_time_t,
# so we take the oldest ticket for each exercise as its last_time_t
last_time_t = old_data['last_time_t'] or oldest_ticket_time_t[exercise]
time_this_period = now - last_time_t
# Avoid divide-by-0 if this is the first time we've seen an exercise
time_last_period = old_data['elapsed_times'].get(exercise, 0.0001)
num_old_tickets_for_exercise = old_data['ticket_counts'][exercise]
num_new_tickets_for_exercise = num_new_tickets[exercise]
(mean, probability) = util.probability(num_old_tickets_for_exercise,
time_last_period,
num_new_tickets_for_exercise,
time_this_period)
print('%s] %s TOTAL %s/%ss; %s-: %s/%ss; m=%.3f p=%.3f'
% (time.strftime('%Y-%m-%d %H:%M:%S %Z'),
exercise,
num_old_tickets_for_exercise, int(time_last_period),
last_time_t,
num_new_tickets_for_exercise, time_this_period,
mean, probability))
if (mean != 0 and probability > 0.9995 and
num_new_tickets_for_exercise > THRESHOLD):
quoted = urllib.quote(exercise.encode("utf-8"))
ka_url = "https://khanacademy.org/e/%s" % quoted
jira_url = "https://khanacademy.atlassian.net/browse/AI-941528?jql=Exercise%%20%%3D%%20%s" % quoted
util.send_to_slack(
"*Elevated bug report rate on exercise `%s`*\n"
"We saw %s in the last %s minutes,"
" while the mean indicates we should see around %s."
" *Probability that this is abnormally elevated: %.4f.*\n"
" Links: <%s|exercise on Khan Academy>, <%s|JIRA tickets>."
% (exercise,
util.thousand_commas(num_new_tickets_for_exercise),
util.thousand_commas(int(time_this_period / 60)),
util.thousand_commas(round(mean, 2)),
probability,
ka_url,
jira_url),
channel='#content-beep-boop')
elapsed_times[exercise] = time_last_period + time_this_period
new_ticket_counts = util.merge_int_dicts(old_data['ticket_counts'],
num_new_tickets)
new_data = {'elapsed_times': elapsed_times,
'ticket_counts': new_ticket_counts,
'last_time_t': now,
}
with open(jira_status_file, 'w') as f:
cPickle.dump(new_data, f)
def main():
try:
zendesk_status_file = util.relative_path("zendesk")
with open(zendesk_status_file) as f:
old_data = cPickle.load(f)
except IOError:
old_data = {"elapsed_time": 0.0001, # avoid a divide-by-0
"ticket_count": 0,
"last_time_t": None,
}
# We compare the number of tickets in the last few minutes against
# the historical average for all time. But we don't start "all
# time" at AD 1, we start it a week ago. Longer than that and it
# takes forever due to quota issues. That's still plenty of
# historical data. :-)
#
# Zendesk seems to wait 5 minutes to update API data :-(, so we
# ask for data that's a bit time-lagged
end_time = int(time.time()) - 300
start_time = old_data['last_time_t']
# Set flag to track if current time period is a weekend. Separate
# ticket_count/elapsed_time stats are kept for weekend vs. weekday
# to improve sensitivity to increases during low-traffic periods
is_weekend = time.localtime().tm_wday in [5, 6]
(num_new_tickets, oldest_ticket_time_t) = num_tickets_between(
start_time or (end_time - 86400 * 7), end_time)
# The first time we run this, we take the starting time to be the
# time of the first bug report.
if start_time is None:
start_time = oldest_ticket_time_t
time_this_period = end_time - start_time
# To handle transition from unsegmented to segmented data, below sets
# the weekend data to mirror the stats from the past 4 months of logs
# to calculate a mean, and shifts all historical data to the weekday
# data points. This will result in some inaccuracy, but the weekend
# data should skew the weekday data only negligably. May cause some
# skewed alerting during the transition period.
# TODO(jacqueline): Remove this transition code after August 2017
if 'elapsed_time' in old_data:
old_data['ticket_count_weekday'] = old_data['ticket_count']
old_data['ticket_count_weekend'] = 555
old_data['elapsed_time_weekday'] = old_data['elapsed_time']
old_data['elapsed_time_weekend'] = 2921756.0001
if is_weekend is True:
ticket_count = old_data['ticket_count_weekend']
elapsed_time = old_data['elapsed_time_weekend']
else:
ticket_count = old_data['ticket_count_weekday']
elapsed_time = old_data['elapsed_time_weekday']
(mean, probability) = util.probability(ticket_count,
elapsed_time,
num_new_tickets,
time_this_period)
print ("%s] TOTAL %s/%ss; %s-: %s/%ss; m=%.3f p=%.3f"
% (time.strftime("%Y-%m-%d %H:%M:%S %Z"),
ticket_count, int(elapsed_time),
start_time,
num_new_tickets, time_this_period,
mean, probability))
handle_alerts(num_new_tickets, time_this_period, mean, probability,
start_time, end_time)
if is_weekend is True:
new_data = {"elapsed_time_weekend": (
old_data["elapsed_time_weekend"] + time_this_period),
"ticket_count_weekend": (
old_data["ticket_count_weekend"] + num_new_tickets),
"elapsed_time_weekday": old_data["elapsed_time_weekday"],
"ticket_count_weekday": old_data["ticket_count_weekday"],
}
else:
new_data = {"elapsed_time_weekend": old_data["elapsed_time_weekend"],
"ticket_count_weekend": old_data["ticket_count_weekend"],
"elapsed_time_weekday": (
old_data["elapsed_time_weekday"] + time_this_period),
"ticket_count_weekday": (
old_data["ticket_count_weekday"] + num_new_tickets),
}
new_data['last_time_t'] = end_time
with open(zendesk_status_file, 'w') as f:
cPickle.dump(new_data, f)