Cox Modeling Using Lifelines

Extending from our notebook on the math and intuition behind the Cox Model let’s do a practical example using real data.

The Data

We’ll use the Telco Customer Churn dataset on Kaggle, which is basically a bunch of client records for a telecom company, where the goal is to predict churn (Churn) and the duration it takes for churn to happen (tenure).

%pylab inline

import pandas as pd
Populating the interactive namespace from numpy and matplotlib

Lot of potentially-useful information here, this notebook does a particularly good job exploring the data.

df = pd.read_csv('https://raw.githubusercontent.com/'
                 'treselle-systems/customer_churn_analysis/'
                 'master/WA_Fn-UseC_-Telco-Customer-Churn.csv')

df.head().T
0 1 2 3 4
customerID 7590-VHVEG 5575-GNVDE 3668-QPYBK 7795-CFOCW 9237-HQITU
gender Female Male Male Male Female
SeniorCitizen 0 0 0 0 0
Partner Yes No No No No
Dependents No No No No No
tenure 1 34 2 45 2
PhoneService No Yes Yes No Yes
MultipleLines No phone service No No No phone service No
InternetService DSL DSL DSL DSL Fiber optic
OnlineSecurity No Yes Yes Yes No
OnlineBackup Yes No Yes No No
DeviceProtection No Yes No Yes No
TechSupport No No No Yes No
StreamingTV No No No No No
StreamingMovies No No No No No
Contract Month-to-month One year Month-to-month One year Month-to-month
PaperlessBilling Yes No Yes No Yes
PaymentMethod Electronic check Mailed check Mailed check Bank transfer (automatic) Electronic check
MonthlyCharges 29.85 56.95 53.85 42.3 70.7
TotalCharges 29.85 1889.5 108.15 1840.75 151.65
Churn No No Yes No Yes

There’s about a 3-1 split between churn and not churn

df['Churn'].value_counts() / len(df)
No     0.73463
Yes    0.26537
Name: Churn, dtype: float64

And the difference in tenure distribution between the two is pretty stark– lot of right-censored data in our No group.

fig, ax = plt.subplots(figsize=(12, 10))

for idx, group in df.groupby('Churn'):
    group['tenure'].hist(ax=ax, bins=20, label=idx)

ax.legend();

png

There’s a lot more here than is illustrative to use, so let’s pare down this dataset a bit.

First, we’ll move customerID into the index as well as drop the TotalCharges variables, as there’s an obvious, uninformative, colinearity between that and tenure.

simple = df.set_index('customerID', drop=True)
simple = simple.drop('TotalCharges', axis=1)

We’ll keep the numeric tenure and MonthlyCharges columns, but here, I’m narrowing down to categorical features that I think might be useful, then dummy-ing them for model consumption.

simple = pd.get_dummies(simple,
                        columns=['gender', 'SeniorCitizen', 'InternetService',
                                 'Contract', 'PaymentMethod', 'Churn'],
                        drop_first=True).select_dtypes('number')

simple.head().T
customerID 7590-VHVEG 5575-GNVDE 3668-QPYBK 7795-CFOCW 9237-HQITU
tenure 1.00 34.00 2.00 45.0 2.0
MonthlyCharges 29.85 56.95 53.85 42.3 70.7
gender_Male 0.00 1.00 1.00 1.0 0.0
SeniorCitizen_1 0.00 0.00 0.00 0.0 0.0
InternetService_Fiber optic 0.00 0.00 0.00 0.0 1.0
InternetService_No 0.00 0.00 0.00 0.0 0.0
Contract_One year 0.00 1.00 0.00 1.0 0.0
Contract_Two year 0.00 0.00 0.00 0.0 0.0
PaymentMethod_Credit card (automatic) 0.00 0.00 0.00 0.0 0.0
PaymentMethod_Electronic check 1.00 0.00 0.00 0.0 1.0
PaymentMethod_Mailed check 0.00 1.00 1.00 0.0 0.0
Churn_Yes 0.00 0.00 1.00 0.0 1.0

Printing some simple statistics, a few things stand out:

  • The gender breakup is about 5050
  • About a third of the pop has DSL internet
  • Over half of the userbase is going month-to-month
simple.describe().T
count mean std min 25% 50% 75% max
tenure 7043.0 32.371149 24.559481 0.00 9.0 29.00 55.00 72.00
MonthlyCharges 7043.0 64.761692 30.090047 18.25 35.5 70.35 89.85 118.75
gender_Male 7043.0 0.504756 0.500013 0.00 0.0 1.00 1.00 1.00
SeniorCitizen_1 7043.0 0.162147 0.368612 0.00 0.0 0.00 0.00 1.00
InternetService_Fiber optic 7043.0 0.439585 0.496372 0.00 0.0 0.00 1.00 1.00
InternetService_No 7043.0 0.216669 0.412004 0.00 0.0 0.00 0.00 1.00
Contract_One year 7043.0 0.209144 0.406726 0.00 0.0 0.00 0.00 1.00
Contract_Two year 7043.0 0.240664 0.427517 0.00 0.0 0.00 0.00 1.00
PaymentMethod_Credit card (automatic) 7043.0 0.216101 0.411613 0.00 0.0 0.00 0.00 1.00
PaymentMethod_Electronic check 7043.0 0.335794 0.472301 0.00 0.0 0.00 1.00 1.00
PaymentMethod_Mailed check 7043.0 0.228880 0.420141 0.00 0.0 0.00 0.00 1.00
Churn_Yes 7043.0 0.265370 0.441561 0.00 0.0 0.00 1.00 1.00

Fitting a Model

So if we take what we’ve got and fit a simple model to it, we can get an easy glimpse at the significance of our features in the p column.

from lifelines import CoxPHFitter

cph = CoxPHFitter().fit(simple, 'tenure', 'Churn_Yes')

cph.print_summary()
model lifelines.CoxPHFitter
duration col 'tenure'
event col 'Churn_Yes'
baseline estimation breslow
number of observations 7043
number of events observed 1869
partial log-likelihood -14102.86
time fit was run 2020-04-06 19:34:49 UTC
coef exp(coef) se(coef) coef lower 95% coef upper 95% exp(coef) lower 95% exp(coef) upper 95% z p -log2(p)
MonthlyCharges -0.03 0.97 0.00 -0.04 -0.03 0.96 0.97 -15.20 <0.005 170.86
gender_Male -0.09 0.91 0.05 -0.18 0.00 0.84 1.00 -1.92 0.05 4.19
SeniorCitizen_1 -0.09 0.92 0.05 -0.19 0.02 0.82 1.02 -1.57 0.12 3.11
InternetService_Fiber optic 1.54 4.66 0.10 1.34 1.74 3.82 5.68 15.22 <0.005 171.46
InternetService_No -1.35 0.26 0.12 -1.59 -1.11 0.20 0.33 -10.91 <0.005 89.58
Contract_One year -1.78 0.17 0.09 -1.95 -1.61 0.14 0.20 -20.50 <0.005 307.77
Contract_Two year -3.52 0.03 0.16 -3.84 -3.20 0.02 0.04 -21.71 <0.005 344.74
PaymentMethod_Credit card (automatic) -0.02 0.98 0.09 -0.20 0.16 0.82 1.17 -0.22 0.82 0.28
PaymentMethod_Electronic check 0.71 2.03 0.07 0.57 0.84 1.76 2.33 9.99 <0.005 75.70
PaymentMethod_Mailed check 0.69 2.00 0.09 0.52 0.87 1.69 2.38 7.91 <0.005 48.48
Concordance 0.84
Log-likelihood ratio test 3100.35 on 10 df
-log2(p) of ll-ratio test inf

I was negative-surprised to see that the Electronic Check and Mailed Check payment methods led to a significant dip in the Surival Curve– me and my Planet Fitness membership are all too familiar with the “out of sight, out of mind” business model. Credit card (automatic), on the other hand, is not significant, because the 4th payment method is Bank transfer (automatic), which basically achieves the same thing

df['PaymentMethod'].value_counts()
Electronic check             2365
Mailed check                 1612
Bank transfer (automatic)    1544
Credit card (automatic)      1522
Name: PaymentMethod, dtype: int64

Plotting

Of course, we can use the model and records from our dataset to predict the Survival Curve

sample = simple.sample(50, random_state=12)

cph.predict_survival_function(sample).plot(figsize=(12, 10),
                                           legend=False);

png

Additionally, we can use plot_covariate_groups() to hold everything equal, save for one attribute, then examine the relative effects of different values on the population.

Here, we can see that having fiber internet tends to make users churn faster.

cph.plot_covariate_groups('InternetService_Fiber optic',
                          values=[0, 1],
                          figsize=(10, 5));

png

But why do you suppose that is?

Comparing Two Populations

My immediate hypothesis is that it’s got something to do with how the client has to pay. Indeed, at all levels of contract, someone with fiber internet pays more. So it could be a fiber thing, or could be a money thing.

simple.groupby(['Contract_Two year', 'Contract_One year',
                'InternetService_Fiber optic'])['MonthlyCharges'].agg(['mean', 'std'])
mean std
Contract_Two year Contract_One year InternetService_Fiber optic
0 0 0 41.278220 17.499133
1 87.021194 11.198021
1 0 45.582923 22.638256
1 98.779499 9.349157
1 0 0 45.927883 26.633123
1 104.571445 9.422629

Let’s pare down to just Fiber users to do some investigation.

fiber = simple[simple['InternetService_Fiber optic'] == 1]
fiber = fiber[['tenure', 'Contract_One year',
               'Contract_Two year', 'MonthlyCharges',
               'gender_Male', 'Churn_Yes']]
cph = CoxPHFitter().fit(fiber, 'tenure', 'Churn_Yes')

cph.print_summary()
model lifelines.CoxPHFitter
duration col 'tenure'
event col 'Churn_Yes'
baseline estimation breslow
number of observations 3096
number of events observed 1297
partial log-likelihood -9055.07
time fit was run 2020-04-06 19:34:49 UTC
coef exp(coef) se(coef) coef lower 95% coef upper 95% exp(coef) lower 95% exp(coef) upper 95% z p -log2(p)
Contract_One year -1.76 0.17 0.11 -1.97 -1.54 0.14 0.21 -16.26 <0.005 195.06
Contract_Two year -3.21 0.04 0.20 -3.59 -2.82 0.03 0.06 -16.22 <0.005 194.20
MonthlyCharges -0.04 0.96 0.00 -0.04 -0.03 0.96 0.97 -13.61 <0.005 137.72
gender_Male -0.14 0.87 0.06 -0.25 -0.03 0.78 0.97 -2.50 0.01 6.33
Concordance 0.79
Log-likelihood ratio test 1361.51 on 4 df
-log2(p) of ll-ratio test 972.71

Actually, it looks like having a higher monthly charge means you’re more likely to stick around.

seed = 42

low = fiber[fiber['MonthlyCharges'] < 80].sample(50, random_state=seed)
hi = fiber[fiber['MonthlyCharges'] > 100].sample(50, random_state=seed)

Well that’s odd.

fig, ax = plt.subplots(figsize=(12, 10))

cph.predict_survival_function(low).plot(ax=ax, c='b', alpha=.5, legend=False)
cph.predict_survival_function(hi).plot(ax=ax, c='r', alpha=.5, legend=False);

from matplotlib.lines import Line2D
custom_lines = [Line2D([0], [0], color='b'),
                Line2D([0], [0], color='r')]

ax.legend(custom_lines, ['Low', 'High']);

png

It’s obvious visually, but if we wanted to say with statistical confidence that the high and low paying populations had different survival curve distributions, we could do so via the built-in logrank_test()

from lifelines.statistics import logrank_test

results = logrank_test(low['tenure'], hi['tenure'],
                       event_observed_A=low['Churn_Yes'],
                       event_observed_B=hi['Churn_Yes'])
results.print_summary()
t_0 -1
null_distribution chi squared
degrees_of_freedom 1
test_name logrank_test
test_statistic p
0 21.90 <0.005
A-ha! Maybe it's the contracts that are holding people locked in. Let's just look at the half of people paying month-to-month ```python from IPython.display import Image not_one = (fiber['Contract_One year'] == 0) not_two = (fiber['Contract_Two year'] == 0) m2m_idx = (not_one & not_two) m2m = fiber[m2m_idx].drop(['Contract_One year', 'Contract_Two year'], axis=1) m2m.head() ```
tenure MonthlyCharges gender_Male Churn_Yes
customerID
9237-HQITU 2 70.70 0 1
9305-CDSKC 8 99.65 0 1
1452-KIOVK 22 89.10 1 0
7892-POOKP 28 104.80 0 1
0280-XJGEX 49 103.70 1 1

Here’s a screengrab of the code. I’m tired of fighting with LaTex.

Image('images/i_hate_latex.PNG')

png

Re-fit and…

cph = CoxPHFitter()

cph.fit(m2m, duration_col='tenure', event_col='Churn_Yes')

cph.print_summary()
model lifelines.CoxPHFitter
duration col 'tenure'
event col 'Churn_Yes'
baseline estimation breslow
number of observations 2128
number of events observed 1162
partial log-likelihood -7955.78
time fit was run 2020-04-06 19:34:50 UTC
coef exp(coef) se(coef) coef lower 95% coef upper 95% exp(coef) lower 95% exp(coef) upper 95% z p -log2(p)
MonthlyCharges -0.04 0.96 0.00 -0.04 -0.03 0.96 0.97 -13.63 <0.005 138.14
gender_Male -0.15 0.86 0.06 -0.26 -0.03 0.77 0.97 -2.50 0.01 6.33
Concordance 0.65
Log-likelihood ratio test 195.83 on 2 df
-log2(p) of ll-ratio test 141.26

Damn it.

cph.plot_covariate_groups('MonthlyCharges', values=[80, 100])
<matplotlib.axes._subplots.AxesSubplot at 0x1d3611169b0>

png

Starting Over

So there’s obviously more at play here than just MonthlyCharges.

Let’s re-pull our original dataset with all of the features, but only for records that are going month-to-month on Fiber internet

m2m_index = m2m.index

a = df.set_index('customerID')
b = a.loc[m2m_index].drop('TotalCharges', axis=1)
c = pd.get_dummies(b, drop_first=True)

m2m_wide = c

Of course, fitting the model with everything but the kitchen sink yields us more than we need.

cph = CoxPHFitter()
cph.fit(m2m_wide, duration_col='tenure', event_col='Churn_Yes')
cph.print_summary()
model lifelines.CoxPHFitter
duration col 'tenure'
event col 'Churn_Yes'
baseline estimation breslow
number of observations 2128
number of events observed 1162
partial log-likelihood -7757.37
time fit was run 2020-04-06 19:34:50 UTC
coef exp(coef) se(coef) coef lower 95% coef upper 95% exp(coef) lower 95% exp(coef) upper 95% z p -log2(p)
SeniorCitizen -0.11 0.90 0.07 -0.24 0.02 0.79 1.02 -1.64 0.10 3.32
MonthlyCharges -0.00 1.00 0.03 -0.06 0.05 0.95 1.06 -0.02 0.98 0.03
gender_Male -0.17 0.85 0.06 -0.28 -0.05 0.75 0.95 -2.79 0.01 7.57
Partner_Yes -0.50 0.61 0.07 -0.64 -0.37 0.53 0.69 -7.25 <0.005 41.09
Dependents_Yes -0.04 0.96 0.09 -0.22 0.15 0.80 1.16 -0.40 0.69 0.54
MultipleLines_Yes -0.46 0.63 0.15 -0.76 -0.16 0.47 0.85 -3.03 <0.005 8.67
OnlineSecurity_Yes -0.74 0.48 0.17 -1.07 -0.41 0.34 0.67 -4.37 <0.005 16.31
OnlineBackup_Yes -0.66 0.52 0.15 -0.96 -0.36 0.38 0.70 -4.26 <0.005 15.55
DeviceProtection_Yes -0.32 0.73 0.15 -0.62 -0.02 0.54 0.98 -2.09 0.04 4.76
TechSupport_Yes -0.59 0.56 0.17 -0.92 -0.25 0.40 0.78 -3.45 <0.005 10.81
StreamingTV_Yes -0.02 0.98 0.29 -0.58 0.55 0.56 1.73 -0.06 0.95 0.07
StreamingMovies_Yes -0.19 0.83 0.29 -0.75 0.37 0.47 1.45 -0.66 0.51 0.98
PaperlessBilling_Yes 0.27 1.30 0.08 0.11 0.42 1.12 1.52 3.35 <0.005 10.25
PaymentMethod_Credit card (automatic) -0.15 0.86 0.12 -0.39 0.09 0.67 1.09 -1.24 0.21 2.23
PaymentMethod_Electronic check 0.57 1.76 0.09 0.39 0.74 1.47 2.10 6.22 <0.005 30.95
PaymentMethod_Mailed check 0.47 1.60 0.13 0.21 0.73 1.24 2.07 3.58 <0.005 11.53
Concordance 0.73
Log-likelihood ratio test 592.65 on 16 df
-log2(p) of ll-ratio test 382.30

Okay, so it looks like it’s less to do with contracts and money, and more to do with additional features. Perhaps a convincing-enough interpretation here is “if the user gets upsold on everything, they’re not probably not shopping around.”

Visually inspecting that p column again, let’s hold out all of the features that look significant and re-fit.

sig_cols = ['gender_Male', 'Partner_Yes', 'MultipleLines_Yes',
            'OnlineSecurity_Yes', 'OnlineBackup_Yes', 'DeviceProtection_Yes',
            'TechSupport_Yes', 'PaperlessBilling_Yes', 'PaymentMethod_Electronic check',
            'PaymentMethod_Mailed check']

cph = CoxPHFitter()
cph.fit(m2m_wide[sig_cols + ['tenure', 'Churn_Yes']], duration_col='tenure', event_col='Churn_Yes')
cph.print_summary()
model lifelines.CoxPHFitter
duration col 'tenure'
event col 'Churn_Yes'
baseline estimation breslow
number of observations 2128
number of events observed 1162
partial log-likelihood -7765.40
time fit was run 2020-04-06 19:34:50 UTC
coef exp(coef) se(coef) coef lower 95% coef upper 95% exp(coef) lower 95% exp(coef) upper 95% z p -log2(p)
gender_Male -0.17 0.84 0.06 -0.29 -0.06 0.75 0.94 -2.92 <0.005 8.18
Partner_Yes -0.52 0.60 0.06 -0.64 -0.39 0.53 0.67 -8.20 <0.005 51.88
MultipleLines_Yes -0.49 0.61 0.06 -0.61 -0.37 0.54 0.69 -7.93 <0.005 48.74
OnlineSecurity_Yes -0.73 0.48 0.09 -0.92 -0.55 0.40 0.58 -7.79 <0.005 47.08
OnlineBackup_Yes -0.66 0.52 0.07 -0.80 -0.53 0.45 0.59 -9.68 <0.005 71.18
DeviceProtection_Yes -0.36 0.70 0.07 -0.49 -0.23 0.61 0.80 -5.38 <0.005 23.66
TechSupport_Yes -0.59 0.55 0.09 -0.78 -0.41 0.46 0.66 -6.31 <0.005 31.70
PaperlessBilling_Yes 0.24 1.27 0.08 0.08 0.39 1.09 1.48 3.01 <0.005 8.60
PaymentMethod_Electronic check 0.60 1.83 0.07 0.46 0.74 1.59 2.10 8.49 <0.005 55.47
PaymentMethod_Mailed check 0.56 1.75 0.12 0.33 0.79 1.39 2.21 4.75 <0.005 18.88
Concordance 0.73
Log-likelihood ratio test 576.60 on 10 df
-log2(p) of ll-ratio test 387.81

Check Assumptions

This looks promising. But did we make a procedural blunder in all of our data transformation?

lifelines gives us an awesome tool that we can use to simply check the Cox Model assumptions

cph.check_assumptions(training_df=m2m_wide[sig_cols + ['tenure', 'Churn_Yes']])
The ``p_value_threshold`` is set at 0.01. Even under the null hypothesis of no violations, some
covariates will be below the threshold by chance. This is compounded when there are many covariates.
Similarly, when there are lots of observations, even minor deviances from the proportional hazard
assumption will be flagged.

With that in mind, it's best to use a combination of statistical tests and visual tests to determine
the most serious violations. Produce visual plots using ``check_assumptions(..., show_plots=True)``
and looking for non-constant lines. See link [A] below for a full example.
null_distribution chi squared
degrees_of_freedom 1
test_name proportional_hazard_test
test_statistic p
DeviceProtection_Yes km 4.88 0.03
rank 6.18 0.01
MultipleLines_Yes km 17.91 <0.005
rank 19.76 <0.005
OnlineBackup_Yes km 3.40 0.07
rank 4.08 0.04
OnlineSecurity_Yes km 4.71 0.03
rank 5.83 0.02
PaperlessBilling_Yes km 0.17 0.68
rank 0.74 0.39
Partner_Yes km 4.38 0.04
rank 4.55 0.03
PaymentMethod_Electronic check km 2.65 0.10
rank 3.54 0.06
PaymentMethod_Mailed check km 9.96 <0.005
rank 11.07 <0.005
TechSupport_Yes km 3.45 0.06
rank 4.42 0.04
gender_Male km 0.69 0.41
rank 0.71 0.40
1. Variable 'MultipleLines_Yes' failed the non-proportional test: p-value is <5e-05. Advice: with so few unique values (only 2), you can include `strata=['MultipleLines_Yes', ...]` in the call in `.fit`. See documentation in link [E] below. 2. Variable 'DeviceProtection_Yes' failed the non-proportional test: p-value is 0.0129. Advice: with so few unique values (only 2), you can include `strata=['DeviceProtection_Yes', ...]` in the call in `.fit`. See documentation in link [E] below. 3. Variable 'PaymentMethod_Mailed check' failed the non-proportional test: p-value is 0.0009. Advice: with so few unique values (only 2), you can include `strata=['PaymentMethod_Mailed check', ...]` in the call in `.fit`. See documentation in link [E] below. --- [A] https://lifelines.readthedocs.io/en/latest/jupyter_notebooks/Proportional%20hazard%20assumption.html [B] https://lifelines.readthedocs.io/en/latest/jupyter_notebooks/Proportional%20hazard%20assumption.html#Bin-variable-and-stratify-on-it [C] https://lifelines.readthedocs.io/en/latest/jupyter_notebooks/Proportional%20hazard%20assumption.html#Introduce-time-varying-covariates [D] https://lifelines.readthedocs.io/en/latest/jupyter_notebooks/Proportional%20hazard%20assumption.html#Modify-the-functional-form [E] https://lifelines.readthedocs.io/en/latest/jupyter_notebooks/Proportional%20hazard%20assumption.html#Stratification Those last 3 points -- `1., 2., 3.` all suggest we try stratifying. But what is that? ### Stratifying Recall the fundamental structure of our Proportional Hazard Model. Everything that we've been working with until now has been centered around finding the right fit for our partial hazard function. However, we can additionally *stratify* our model to allow for *multiple* baseline hazard functions. This should track, intuitively, as we'd generally expect the *overall shape* of survival curves to vary between populations-- the linear scaling can only achieve so much, after all. ```python from IPython.display import Image Image('images/lifelines_cox_eq.PNG') ``` ![png](cox_lifelines_51_0.png) And implementing it is as wasy as populating the `strata` argument ```python cph = CoxPHFitter() cph.fit(m2m_wide[sig_cols + ['tenure', 'Churn_Yes']], duration_col='tenure', event_col='Churn_Yes', strata=['MultipleLines_Yes', 'DeviceProtection_Yes', 'PaymentMethod_Mailed check']) cph.print_summary() ```
model lifelines.CoxPHFitter
duration col 'tenure'
event col 'Churn_Yes'
strata [MultipleLines_Yes, DeviceProtection_Yes, Paym...
baseline estimation breslow
number of observations 2128
number of events observed 1162
partial log-likelihood -6049.24
time fit was run 2020-04-06 19:34:50 UTC
coef exp(coef) se(coef) coef lower 95% coef upper 95% exp(coef) lower 95% exp(coef) upper 95% z p -log2(p)
gender_Male -0.17 0.85 0.06 -0.28 -0.05 0.75 0.95 -2.81 <0.005 7.67
Partner_Yes -0.52 0.59 0.06 -0.65 -0.40 0.52 0.67 -8.17 <0.005 51.53
OnlineSecurity_Yes -0.73 0.48 0.09 -0.92 -0.55 0.40 0.58 -7.80 <0.005 47.18
OnlineBackup_Yes -0.67 0.51 0.07 -0.81 -0.54 0.45 0.58 -9.71 <0.005 71.63
TechSupport_Yes -0.60 0.55 0.09 -0.79 -0.42 0.46 0.66 -6.35 <0.005 32.07
PaperlessBilling_Yes 0.24 1.27 0.08 0.08 0.39 1.08 1.48 2.98 <0.005 8.46
PaymentMethod_Electronic check 0.60 1.83 0.07 0.46 0.74 1.59 2.10 8.43 <0.005 54.74
Concordance 0.69
Log-likelihood ratio test 416.12 on 7 df
-log2(p) of ll-ratio test 282.63

Re-checking, we can see that lifelines doesn’t hate the work we did, this time~

cph.check_assumptions(m2m_wide[sig_cols + ['tenure', 'Churn_Yes']])
Proportional hazard assumption looks okay.

Of course, this now means that our visual interpretability goes down

m2m_sample = m2m_wide[sig_cols + ['tenure', 'Churn_Yes']].sample(50)

cph.predict_cumulative_hazard(m2m_sample).plot(legend=False,
                                               figsize=(15, 10));

png

and while lifelines allows us to inspect the baseline survival curves, for the various strata

cph.baseline_survival_.plot(figsize=(15, 8))
plt.gcf().suptitle('{}'.format(cph.strata));

png

This gets us to the fundamental tension of “model accuracy” and “model interpretability” and is one of the many corners of “Data Science is more of an art than a science”