sessions.py

#from https://www.dataiku.com/learn/guide/code/reshaping_data/sessionization.html
import dataiku
import pandas as pd
from datetime import timedelta

# define treshold value
T = timedelta(seconds=30*60)

# load dataset
toy_data = dataiku.Dataset("toy_data").get_dataframe()

# add a column containing previous timestamp
toy_data =  pd.concat([toy_data,
                       toy_data.groupby('user_id').transform(lambda x:x.shift(1))]
                      ,axis=1)
toy_data.columns = ['user_id','mytimestamp','prev_mytimestamp']

# create the new session column 
toy_data['new_session'] = ((toy_data['mytimestamp'] 
                            - toy_data['prev_mytimestamp'])>=T).astype(int)

# create the session_id
toy_data['increment'] = toy_data.groupby("user_id")['new_session'].cumsum()
toy_data['session_id'] = toy_data['user_id'].astype(str) + '_' 
                                + toy_data['increment'].astype(str)

# to get the same result as with hive/postgresql
toy_data = toy_data.sort(['user_id','mytimestamp'])