

In [2]:

    
# Choose range of user IDs

ids = list(range(5000,6001))

id_list = []

# Repeat each user ID a random number of times based on a Gaussian distribution
for i in ids:
    y = np.repeat(i, int(random.gauss(25,8))).tolist()
    id_list.append(y)
    
id_list = [item for sublist in id_list for item in sublist]

date_list = []

# Add a random date to each user ID
for i in id_list:
    year = 2016
    month = random.randint(1,12)
    day = random.randint(1,28)
    hour = random.randint(0,23)
    minute = random.randint(0,59)
    second = random.randint(0,59)
    date = datetime.datetime(year, month, day, hour, minute, second)
    
    date_list.append(date)

# Convert each list to Series/DataFrame
id_series = pd.Series(id_list, name='user_id')
date_series = pd.Series(date_list, name='event_timestamp')

df = pd.concat([id_series,date_series], axis=1)

df = df.sort_values('event_timestamp').reset_index(drop=True)

# Create additional columns in DF, with each having the 'login' event_type
    
df['lat'] = 0.0
df['lon'] = 0.0
df['event_type'] = "login"

# Create random latitude and longitude coordinates for each event, centered on Chicago
def lat_lon(group):
    
    x0 = 41.8781
    y0 = 87.6298
    
    for i in range(0,len(group)):
        
        # Ensure that each successive lat/lon does not deviate too far from the initial
        radius = random.randint(2000,50000)
        rad_deg = radius/111300
        r = rad_deg
        
        u = float(random.uniform(0.0,1.0))
        v = float(random.uniform(0.0,1.0))
        w = r * math.sqrt(u)
        t = 2*math.pi*v
        x = w*math.cos(t)
        y = w*math.sin(t)
        group['lat'].iloc[i] = x + x0
        group['lon'].iloc[i] = y + y0
        
    return group

df = df.groupby('user_id').apply(lat_lon)

df = df.sort_values(by=['user_id','event_timestamp']).reset_index(drop=True)

# Function for creating a random number of events for every 'login' event
def add_events(group):
    
    event_num = random.randint(1,10)

    idlist = np.repeat(group['user_id'],event_num).tolist()

    # Increment timestamp randomly, with 5-15 min as boundaries
    timestamp = group['event_timestamp'].iloc[0]

    timestamplist = []

    for i in range(0,event_num):
        timestamp += datetime.timedelta(minutes=random.randint(5,15))
        timestamplist.append(timestamp)
    
    # Insert the same lat/lon for each new as the 'login' event
    latlist = np.repeat(group['lat'], event_num).tolist()
    lonlist = np.repeat(group['lon'], event_num).tolist()
    
    # For each new event, probabilistically choose one of the 3 additional events
    events = ['level', 'buy_coins', 'megapack']
    eventlist = np.random.choice(events, event_num, p=[0.85,0.1,0.05])

    # Create new dataframe from the new data
    d = {'user_id': idlist, 'event_timestamp': timestamplist, 'lat': latlist, 'lon': lonlist, 'event_type': eventlist}
    
    eventdf = pd.DataFrame(d)
    
    # Append the newly-generated data to the initial group
    group = group.append(eventdf, ignore_index=True)
    
    return group

# Apply each new group to 
df = df.groupby(['user_id','event_timestamp']).apply(add_events)

# Function to increment each occurrence of the 'level' event_type
def level_up(group):
    level=1
    for i in range(0,len(group)):
        if group['event_type'].iloc[i] == "level":
            group['event_type'].iloc[i] = "level_" + str(level)
            level+=1

    return group

df = df.groupby(['user_id']).apply(level_up)

df = df[['user_id', 'event_timestamp', 'lat','lon','event_type']]

df.to_csv("test_user_data.csv", index=False)