Jupyter

JupyterLab в учебном маршруте

churn_features.ipynb

Lesson 4 · Churn features

Проверьте признаки перед обучением модели. Датасет уже лежит в рабочей папке урока.

In [1]

import pandas as pd
import matplotlib.pyplot as plt

customers = pd.read_csv('customers.csv')
customers[['tenure_months', 'tickets_30d', 'nps', 'churned']].head()

   tenure_months  tickets_30d   nps  churned
0              4            3   6.0        1
1             18            0   9.0        0
2              9            2   7.0        0
3             31            1   8.0        0
4              2            5   3.0        1

5 rows × 4 columns

In [2]

features = customers.assign(
    ticket_rate=customers['tickets_30d']
    / customers['tenure_months'].clip(lower=1)
)
features[['tenure_months', 'ticket_rate', 'nps', 'churned']].corr().round(3)

               tenure_m  ticket_rate      nps   churned
tenure_m         1.000       -0.412   0.186    -0.341
ticket_rate     -0.412        1.000  -0.274     0.503
nps              0.186       -0.274   1.000    -0.298
churned         -0.341        0.503  -0.298     1.000

4 × 4

In [3]

tenure_bin = pd.cut(
    customers['tenure_months'],
    bins=[0, 6, 12, 24, 120],
    labels=['0–6m', '6–12m', '12–24m', '24m+'],
)
rate = customers.groupby(tenure_bin, observed=True)['churned'].mean() * 100
rate.plot(kind='bar', color='#0ea5e9', rot=0)
plt.ylabel('churn rate, %')
plt.title('Churn rate by tenure')
plt.tight_layout()

28.4

0–6m

19.1

6–12m

11.6

12–24m

6.2

24m+

Mode: CommandPython 3.11 | IdleMock · no remote kernel