zulip/zerver/lib/statistics.py

124 lines
4.7 KiB
Python

# -*- coding: utf-8 -*-
from __future__ import absolute_import
from __future__ import division
from zerver.models import UserProfile, UserActivity, UserActivityInterval, Message
from django.utils.timezone import utc
from datetime import timedelta
from itertools import chain
from six.moves import range
import six
def median(data):
data = sorted(data)
size = len(data)
if size % 2 == 1:
return data[size//2]
else:
before = size//2 - 1
after = size//2
return (data[before] + data[after]) / 2.0
users_who_sent_query = Message.objects.select_related("sender") \
.exclude(sending_client__name__contains="mirror") \
.exclude(sending_client__name__contains="API")
def active_users():
# Return a list of active users we want to count towards various
# statistics.
return UserProfile.objects.filter(is_bot=False, is_active=True).select_related()
def users_who_sent_between(begin, end):
sender_objs = users_who_sent_query.filter(pub_date__gt=begin, pub_date__lt=end) \
.values("sender__id")
return set(s["sender__id"] for s in sender_objs)
def users_who_sent_ever():
return set(s["sender__id"] for s in users_who_sent_query.values("sender__id"))
def active_users_to_measure():
senders = users_who_sent_ever()
return [u for u in active_users() if u.id in senders]
def active_users_who_sent_between(begin, end):
senders = users_who_sent_between(begin, end)
return [u for u in active_users() if u.id in senders]
# Return the amount of Zulip usage for this user between the two
# given dates
def seconds_usage_between(user_profile, begin, end):
intervals = UserActivityInterval.objects.filter(user_profile=user_profile, end__gte=begin, start__lte=end)
duration = timedelta(0)
for interval in intervals:
start = max(begin, interval.start)
finish = min(end, interval.end)
duration += finish-start
return duration
# Return a list of how many seconds each user has been engaging with the app on a given day
def seconds_active_during_day(day):
begin_day = day.replace(hour=0, minute=0, second=0, microsecond=0, tzinfo=utc)
end_day = day.replace(hour=23, minute=59, second=59, microsecond=0, tzinfo=utc)
active_users = active_users_to_measure()
return [seconds_usage_between(user, begin_day, end_day).total_seconds() for user in active_users]
def users_active_nosend_during_day(day):
begin_day = day.replace(hour=0, minute=0, second=0, microsecond=0, tzinfo=utc)
end_day = day.replace(hour=23, minute=59, second=59, microsecond=0, tzinfo=utc)
active_users = active_users_to_measure()
today_senders = users_who_sent_between(begin_day, end_day)
today_users = []
for user_profile in active_users:
intervals = UserActivityInterval.objects.filter(user_profile=user_profile,
end__gte=begin_day,
start__lte=end_day)
if len(intervals) != 0:
today_users.append(user_profile)
return [u for u in today_users if not u.id in today_senders]
def calculate_stats(data, all_users):
if len(data) == 0:
return {"# data points": 0}
active_user_count = len([x for x in data if x > 1])
mean_data = sum(data) // active_user_count
median_data = median([x for x in data if x > 1])
return {'active users': active_user_count,
'total users': len(all_users),
'mean': str(timedelta(seconds=mean_data)),
'median': str(timedelta(seconds=median_data)),
'# data points': len(data)}
# Return an info dict {mean: , median} containing the mean/median seconds users were active on a given day
def activity_averages_during_day(day):
users_to_measure = active_users_to_measure()
seconds_active = seconds_active_during_day(day)
return calculate_stats(seconds_active, all_users=users_to_measure)
# Returns an info dict {mean: , median} with engagement numbers for all users according
# to active_users_to_measure.
def activity_averages_between(begin, end, by_day=True):
seconds_active = {}
users_to_measure = active_users_to_measure()
for i in range((end - begin).days):
day = begin + timedelta(days=i)
# Ignore weekends
if day.weekday() in [5, 6]:
continue
seconds_active[day] = seconds_active_during_day(day)
if by_day:
return dict((day, calculate_stats(values, all_users=users_to_measure))
for day, values in six.iteritems(seconds_active))
else:
return calculate_stats(list(chain.from_iterable(seconds_active.values())),
all_users=users_to_measure)