VladimirKazantsevPM / lifetimes

Customer LTV in Python

lifetimes

Measuring customer lifetime value is hard. lifetimes makes it easy.

Quickstart

Example using transactional data:

from matplotlib import pyplot as plt
from lifetimes.estimation import BetaGeoFitter
from lifetimes.utils import summary_data_from_transaction_data

# transactions representing customers' shopping history.
transactions_A = pd.DataFrame([
    {'id': 1, 'date': '2015-01-01'},
    {'id': 1, 'date': '2015-01-05'},
    {'id': 1, 'date': '2015-01-08'},
    {'id': 2, 'date': '2015-01-04'},
    {'id': 2, 'date': '2015-01-30'},
    {'id': 3, 'date': '2015-01-10'},
    {'id': 3, 'date': '2015-02-05'},
    {'id': 3, 'date': '2015-02-06'},
])
data_A = summary_data_from_transaction_data(transactions_A, 'id', 'date')

transactions_B = pd.DataFrame([
    {'id': 1, 'date': '2015-01-01'},
    {'id': 1, 'date': '2015-02-05'},
    {'id': 2, 'date': '2015-02-01'},
    {'id': 2, 'date': '2015-02-05'},
    {'id': 3, 'date': '2015-01-16'},
    {'id': 3, 'date': '2015-02-05'},
    {'id': 3, 'date': '2015-02-06'},
    {'id': 3, 'date': '2015-02-07'},
])
data_B = summary_data_from_transaction_data(transactions_B, 'id', 'date')


bgf = BetaGeoFitter()
bgf.fit(data_A['frequency'], data_A['recency'], data_A['cohort'])
ax = bgf.plot(label='group A')

bgf.fit(data_B['frequency'], data_B['recency'], data_B['cohort'])
ax = bgf.plot(label='group B')
plt.legend(loc='lower right')

Example to assess model fit

%matplotlib
import pandas as pd
from lifetimes.estimation import BetaGeoFitter

df = pd.read_csv('lifetimes/datasets/cdnow_customers.csv', sep='\s+', index_col=[0])
bg = BetaGeoFitter()
bg.fit(df['x'], df['t_x'], df['T'] )

bg.plot_period_transactions()