# In[ ]:
import matplotlip
from matplotlib import pyplot as plt
import lifelines
from lifelines import KaplanMeierFitter #survival analysis library
from lifelines.statistics import logrank_test #survival statistical testing
from lifelines import CoxPHFitter
df['churn'] = df1.fuga
cph = CoxPHFitter()
cph.fit(df,
duration_col=ypd1['enddt'],
event_col=ypd1['FUGA'],
show_progress=True)
cph.print_summary()
cph.plot()
# In[ ]:
df_2 = df.drop(['enddt', 'FUGA'], axis=1)
cph.predict_partial_hazard(df_2)
cph.predict_survival_function(df_2, times=[5., 25., 50.])
cph.predict_median(X)
kmf = KaplanMeierFitter()
T = df['time_to_fuga'] #duration
C = df['churn'] #censorship - 1 if death/churn is seen, 0 if censored
# -*- coding: utf-8 -*-
# cox regression
if __name__ == "__main__":
import pandas as pd
import time
import numpy as np
from lifelines import CoxPHFitter
from lifelines.datasets import load_rossi
df = load_rossi()
df = pd.concat([df] * 20)
# df = df.reset_index()
# df['week'] = np.random.exponential(1, size=df.shape[0])
cp = CoxPHFitter()
cp.fit(df, duration_col="week", event_col="arrest", batch_mode=True)
start_time = time.time()
print(cp.predict_median(df))
print("--- %s seconds ---" % (time.time() - start_time))
cp.print_summary(4)
# now looking at the RemainingValue column we can see which customers would most affect our bottom line.
# Great, so we know which customers have the highest risk of churn and when they are likely to, but what can we do?
# lets take a look at our coefficients from earlier.
# we can see that the features that impact survival positively are 'Contract_One year', 'Contract_Two year',
# 'PaymentMethod_Bank transfer (automatic)', 'PaymentMethod_Credit card (automatic)'. Beyond these the results are
# insignificant. Lets compare customers with the features to understand the best place to spend money.
upgrades = ['Contract_One year',
'Contract_Two year',
'PaymentMethod_Bank transfer (automatic)',
'PaymentMethod_Credit card (automatic)']
results_dict = {}
for customer in tqdm(values.index):
actual = data.loc[[customer]]
change = data.loc[[customer]]
results_dict[customer] = [cph.predict_median(actual)]
for upgrade in upgrades:
change[upgrade] = 1 if list(change[upgrade]) == [0] else 0
results_dict[customer].append(cph.predict_percentile(actual, p=likelihood_cutoff))
change[upgrade] = 1 if list(change[upgrade]) == [0] else 0
results_df = pd.DataFrame(results_dict).T
results_df.columns = ['baseline'] + upgrades
actions = values.join(results_df).drop([likelihood_cutoff], axis=1)
# now we can calculate the difference between applying different features from the baseline
actions['CreditCard Diff'] = (
actions['PaymentMethod_Credit card (automatic)'] -
actions['baseline']
) * actions['MonthlyCharges']
actions['BankTransfer Diff'] = (
actions['PaymentMethod_Bank transfer (automatic)'] -
weight_delays=weight_delays)
#Create and train Cox Proportional Hazards model
cph = CoxPHFitter()
if weight_delays != None:
cph.fit(n_train,
duration_col='Target',
weights_col='Weights',
show_progress=True)
else:
cph.fit(n_train, duration_col='Target', show_progress=True)
cph.print_summary()
#Quantify predictions
y_test = n_test.Target.values
y_pred = cph.predict_median(n_test).values.T[0]
#get mean average error for last 100 days
mae_test = sutils.plot_mae_vs_y_true(y_pred, y_test, 100)
#Plot results for presentation slides if necessary
if plot_result:
#Define times at which to plot CDF
times = np.arange(100).tolist()
#Plot lines delimiting an interval in time and corresponding values of the CDF
low_x = 60
high_x = 80
low_y = 1 - cph.predict_survival_function(n_test[70:71], [low_x])
high_y = 1 - cph.predict_survival_function(n_test[70:71], [high_x])