def most_recent_actions():
"""
Returns the most recent events for a user, primarily used in featurizing
"""
events = session.query(Users).join(Event).add_entity(Event).group_by(Users.id).order_by(Event.date.desc()).subquery()
return query_to_df(session, events)
def most_recent_visits():
"""
Returns the most recent visits for a user, primarily used in featurizing
"""
visits = session.query(Users).join(Visit).add_entity(Visit).group_by(
Users.id).order_by(Visit.date.desc())
return query_to_df(session, visits)
def most_recent_actions():
"""
Returns the most recent events for a user, primarily used in featurizing
"""
events = session.query(Users).join(Event).add_entity(Event).group_by(Users.id).order_by(Event.date.desc()).subquery()
return query_to_df(session, events)
def user_visited_in_last_k_days(threshold):
now = datetime.utcnow()
days = now - timedelta(days=threshold)
"""
Only grab the logins that occurred in the last 90 days
"""
most_recent_user_visits = session.query(Users.id,Visit.date,Users.Campaign_ID).join(Visit,Users.id == Visit.user_id).order_by(Visit.date.desc()).group_by(Users.id).filter(Users.Campaign_ID == 'TW')
df = query_to_df(session,most_recent_user_visits)
df = df.drop('Users_Campaign_ID',axis=1)
df['churned'] = df['visit_date'].apply(lambda x : x < days)
df = df.reset_index()
print type(df)
return df
def user_visited_in_last_k_days(threshold):
now = datetime.utcnow()
days = now - timedelta(days=threshold)
"""
Only grab the logins that occurred in the last 90 days
"""
most_recent_user_visits = session.query(Users.id,Visit.date,Users.Campaign_ID).join(Visit,Users.id == Visit.user_id).order_by(Visit.date.desc()).group_by(Users.id).filter(Users.Campaign_ID == 'TW')
df = query_to_df(session,most_recent_user_visits)
df = df.drop('Users_Campaign_ID',axis=1)
df['churned'] = df['visit_date'].apply(lambda x : x < days)
df = df.reset_index()
print type(df)
return df
from pandas import DataFrame
from util import query_to_df
from util import campaign_to_num, event_to_num, transform_column, hist_and_show, vectorize
db = create_engine('sqlite:///forjar.db')
metadata = MetaData(db)
Session = sessionmaker(bind=db)
session = Session()
"""
Counts the users by campaign id
"""
user_dist = session.query(Users)
user_df = query_to_df(session, user_dist)
transform_column(user_df, 'Users_Campaign_ID', campaign_to_num.get)
hist_and_show(user_df, 'Users_Campaign_ID')
q = session.query(Users.Campaign_ID, Event.Type, Users.id,
Event.User_Id).filter(Event.Type == 'bought')
d = query_to_df(session, q)
print d.columns
transform_column(d, 'Users_Campaign_ID', campaign_to_num.get)
"""
Show the counts for the event types
"""
transform_column(d, 'Event_Type', event_to_num.get)
hist_and_show(d, 'Users_Campaign_ID')
db = create_engine('sqlite:///forjar.db')
metadata = MetaData(db)
Session = sessionmaker(bind=db)
session = Session()
"""
Counts the users by campaign id
"""
user_dist = session.query(Users).join(Visit,Users.id == Visit.user_id).add_entity(Visit)
user_df = query_to_df(session,user_dist)
print user_df.columns
user_df = user_df.groupby(['Users_id','visit_date'])
user_logins = {}
def add_to_logins(group):
group = group.apply(pd.to_datetime)
print group.as_matrix()
return rolling_mean(group.as_matrix(),window='30d')
user_df = user_df.aggregate(add_to_logins)
print user_df
metadata = MetaData(db)
Session = sessionmaker(bind=db)
session = Session()
"""
We only want events and users such that the user bought an item.
We count bought as $1 of revenue for simplicity.
"""
q = session.query(Users.Campaign_ID, Event.Type).filter(Event.Type == 'bought')
"""
Print out the counts by name.
This is a way of showing how to aggregate by campaign ids.
"""
df = query_to_df(session, q)
"""
Basic statistics collecting; calculate the sum for each row.
"""
sum = 0
for campaign_id in campaign_to_num.keys():
rows = num_rows(df.groupby('Users_Campaign_ID').get_group(campaign_id))
sum = sum + rows
print 'id was ' + campaign_id + ' ' + str(rows)
"""
Print the percentage of buy actions for each campaign id
"""
for campaign_id in campaign_to_num.keys():
rows = num_rows(df.groupby('Users_Campaign_ID').get_group(campaign_id))
rows = rows / sum
print 'id was ' + campaign_id + ' and percent of customers bought was ' + str(
Session = sessionmaker(bind=db)
session = Session()
"""
We only want events and users such that the user bought an item.
We count bought as $1 of revenue for simplicity.
"""
q = session.query(Users.Campaign_ID,Event.Type).filter(Event.Type == 'bought')
"""
Print out the counts by name.
This is a way of showing how to aggregate by campaign ids.
"""
df = query_to_df(session,q)
"""
Basic statistics collecting; calculate the sum for each row.
"""
sum = 0
for campaign_id in campaign_to_num.keys():
rows = num_rows(df.groupby('Users_Campaign_ID').get_group(campaign_id))
sum = sum + rows
print 'id was ' + campaign_id + ' ' + str(rows)
"""
Print the percentage of buy actions for each campaign id
"""
for campaign_id in campaign_to_num.keys():
rows = num_rows(df.groupby('Users_Campaign_ID').get_group(campaign_id))
rows = rows / sum
def most_recent_visits():
"""
Returns the most recent visits for a user, primarily used in featurizing
"""
visits = session.query(Users).join(Visit).add_entity(Visit).group_by(Users.id).order_by(Visit.date.desc())
return query_to_df(session, visits)