-
Notifications
You must be signed in to change notification settings - Fork 0
/
Kaplan Meier Survival.py
235 lines (179 loc) · 11.8 KB
/
Kaplan Meier Survival.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
import os; os.chdir('C:/Users/marvin/Desktop/SQL BigQuery')
################################################### SQL BigQuery ########################################################
import numpy as np ; import pandas as pd ; from google.cloud import bigquery
client = bigquery.Client.from_service_account_json(
json_credentials_path='data_pipeline-bbc9aec8eae9.json',
project='data_pipeline')
#Pour chaque visiteur on calcule la durée entre sa premiere visite et son premier achat
#s'il en a effectué et la durée entre sa premiere et sa derniere visite sinon
#On indique également le device sur lequel il a effectué la transaction ou sa dernière visite sinon.
#On code 1 s'il a effectué une transaction 0 sinon.
query = """
WITH
visit AS (
SELECT fullvisitorid, MIN(date) AS date_first_visit, MAX(date) AS date_last_visit
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*` GROUP BY fullvisitorid),
device_visit AS (
SELECT DISTINCT fullvisitorid, date, device.deviceCategory
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`),
transactions AS (
SELECT fullvisitorid, MIN(date) AS date_transactions, 1 AS transaction
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*` AS ga,
UNNEST(ga.hits) AS hits WHERE hits.transaction.transactionId IS NOT NULL GROUP BY fullvisitorid),
device_transactions AS (
SELECT DISTINCT fullvisitorid, date, device.deviceCategory
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*` AS ga,
UNNEST(ga.hits) AS hits WHERE hits.transaction.transactionId IS NOT NULL),
visits_transactions AS (
SELECT visit.fullvisitorid, date_first_visit, date_transactions, date_last_visit ,
device_visit.deviceCategory AS device_last_visit,
device_transactions.deviceCategory AS device_transaction,
IFNULL(transactions.transaction,0) AS transaction
FROM visit LEFT JOIN transactions ON visit.fullvisitorid = transactions.fullvisitorid
LEFT JOIN device_visit ON visit.fullvisitorid = device_visit.fullvisitorid
AND visit.date_last_visit = device_visit.date
LEFT JOIN device_transactions ON visit.fullvisitorid = device_transactions.fullvisitorid
AND transactions.date_transactions = device_transactions.date ),
mortality_table AS (
SELECT fullvisitorid, date_first_visit,
CASE WHEN date_transactions IS NULL THEN date_last_visit ELSE date_transactions END AS date_event,
CASE WHEN device_transaction IS NULL THEN device_last_visit ELSE device_transaction END AS device, transaction
FROM visits_transactions )
SELECT CONCAT("ID",fullvisitorid) AS fullvisitorid,
DATE_DIFF(PARSE_DATE('%Y%m%d',date_event),PARSE_DATE('%Y%m%d', date_first_visit),DAY) AS time,
transaction, device FROM mortality_table"""
query_results = client.query(query) ; query_results = query_results.result()
#Résutats de la reqête
fullvisitorid = [] ; time = [] ; device = [] ; transaction = []
for row in query_results:
fullvisitorid.append(row[0])
time.append(row[1])
transaction.append(row[2])
device.append(row[3])
BigQuery_table = {"fullvisitorid":fullvisitorid, "time":time, "transaction":transaction, "device":device}
BigQuery_table = pd.DataFrame(BigQuery_table)
########################################### Analyse de survie de Kaplan Meier ###########################################
#On importe le jeu de donnée de la table BigQuery
col = list(BigQuery_table.columns); del col[0]
survival_data = pd.DataFrame(np.c_[BigQuery_table.iloc[:,1:4]], columns = col, index = BigQuery_table['fullvisitorid'])
#On crée le modèle
from lifelines import KaplanMeierFitter ; import matplotlib.pyplot as plt
kmf = KaplanMeierFitter() ; kmf.fit(survival_data['time'],survival_data['transaction'])
#Résultats
event_table = kmf.event_table #table des évenements
kmf.predict(11) #prediction d'etre en vie à 11 jours
survival_probability = kmf.survival_function_ #table des probabilités de survie
#Courbe de la fonction de survie de Kaplan Meier
kmf.plot(); plt.xlabel("time in days"); plt.ylabel("survival probability"); plt.title("Kaplan Meier Survival Function")
#Courbe plus lisible
plt.figure(figsize=(15,10)) ; kmf.plot() ; plt.xlabel('time in days') ; plt.ylabel('survival probability')
plt.title("Kaplan Meier Survival Function") ; plt.ylim([0.45,1])
########################################## segmentaion & test du log-rank ################################################
#On divise le jeu de données en fonction des segment
desktop = survival_data.query("device == 'desktop'")
mobile = survival_data.query("device == 'mobile'")
tablet = survival_data.query("device == 'tablet'")
#On crée les modèle
kmf_desktop = KaplanMeierFitter() ; kmf_mobile = KaplanMeierFitter() ; kmf_tablet = KaplanMeierFitter()
kmf_desktop.fit(desktop['time'],desktop['transaction'])
kmf_mobile.fit(mobile['time'],mobile['transaction'])
kmf_tablet.fit(tablet['time'],tablet['transaction'])
#Courbes de survie en fonction des segments
plt.figure(figsize=(15,10)) ; plt.ylim([0.45,1])
kmf_desktop.plot(label='desktop') ; kmf_mobile.plot(label='mobile') ; kmf_tablet.plot(label='tablet')
plt.xlabel("time in days") ; plt.ylabel("survival probability") ; plt.title("Kaplan Meier Survival Functions")
#test du log-rank
from lifelines.statistics import multivariate_logrank_test
log_rank_test = multivariate_logrank_test(survival_data['time'], survival_data['device'], survival_data['transaction'])
log_rank_test.summary
#les différence de survie sont significative car la p-valeur est inferieure à 5%
##################################################### Modele de Cox ######################################################
#On calucule l'impact des différentes modalités sur le risque absolu de décès (effectuer une transaction)
from lifelines import CoxPHFitter
#On recode en 2 la modalité que l'on souhaite analyser et 1 les autres
survival_data = pd.DataFrame(np.c_[BigQuery_table.iloc[:,1:4]], columns = col, index = BigQuery_table['fullvisitorid'])
survival_data["device"] = survival_data.device.replace({"desktop":2,"mobile":1, "tablet":1})
cph = CoxPHFitter() ; cph.fit(survival_data,"time","transaction") ; cox_reg = cph.summary ; cox_reg
#le rélultat est significatif car la p-valeure est inferieure à 5%
#La valeure exp(coef) mesure l'impact de la modalité sur le risque absolue de décès
#Si exp(coef) = 1 il n'y a pas d'effet sur le risque absolue de decéder
#Si exp(coef) < 1 avoir cette modalité réduit le risque de décéder
#exp(coef) > 1 augmente le risque de le risque de décéder
#Ici être sur la version sur desktop multiplie le risque de décéder par 3.128524 6 soit une augmentation de 213%
survival_data = pd.DataFrame(np.c_[BigQuery_table.iloc[:,1:4]], columns = col, index = BigQuery_table['fullvisitorid'])
survival_data["device"] = survival_data.device.replace({"desktop":1,"mobile":2, "tablet":1})
cph = CoxPHFitter() ; cph.fit(survival_data,"time","transaction") ; cox_reg = cph.summary ; cox_reg
#le rélultat est significatif car la p-valeure est inferieure à 5%
#Ici être sur la version sur mobile multiplie le risque de décéder par 0.316285 6 soit une réduction de 68%
survival_data = pd.DataFrame(np.c_[BigQuery_table.iloc[:,1:4]], columns = col, index = BigQuery_table['fullvisitorid'])
survival_data["device"] = survival_data.device.replace({"desktop":1,"mobile":1, "tablet":2})
cph = CoxPHFitter() ; cph.fit(survival_data,"time","transaction") ; cox_reg = cph.summary ; cox_reg
#le rélultat est significatif car la p-valeure est inferieure à 5%
##Ici être sur la version sur tablet multiplie le risque de décéder par 0.469092 6 soit une réduction de 53%
#Prédiction de la survie d'un individus
d_data = survival_data.iloc[0:5,:]
cph.predict_survival_function(d_data).plot()
###################################################### Retention #########################################################
#On calcule maintenant la durée entre sa premiere visite et son dernier achat
query = """
WITH
visit AS (
SELECT fullvisitorid, MIN(date) AS date_first_visit, MAX(date) AS date_last_visit
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*` GROUP BY fullvisitorid),
device_visit AS (
SELECT DISTINCT fullvisitorid, date, device.deviceCategory
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*`),
transactions AS (
SELECT fullvisitorid, MAX(date) AS date_transactions, 1 AS transaction
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*` AS ga, UNNEST(ga.hits) AS hits
WHERE hits.transaction.transactionId IS NOT NULL GROUP BY fullvisitorid),
device_transactions AS (
SELECT DISTINCT fullvisitorid, date, device.deviceCategory
FROM `bigquery-public-data.google_analytics_sample.ga_sessions_*` AS ga, UNNEST(ga.hits) AS hits
WHERE hits.transaction.transactionId IS NOT NULL),
visits_transactions AS (
SELECT visit.fullvisitorid, date_first_visit, date_transactions, date_last_visit ,
device_visit.deviceCategory AS device_last_visit, device_transactions.deviceCategory AS device_transaction,
IFNULL(transactions.transaction,0) AS transaction
FROM visit LEFT JOIN transactions ON visit.fullvisitorid = transactions.fullvisitorid
LEFT JOIN device_visit ON visit.fullvisitorid = device_visit.fullvisitorid
AND visit.date_last_visit = device_visit.date
LEFT JOIN device_transactions ON visit.fullvisitorid = device_transactions.fullvisitorid
AND transactions.date_transactions = device_transactions.date ),
mortality_table AS (
SELECT fullvisitorid, date_first_visit,
CASE WHEN date_transactions IS NULL THEN date_last_visit ELSE date_transactions END AS date_event,
CASE WHEN device_transaction IS NULL THEN device_last_visit ELSE device_transaction END AS device, transaction
FROM visits_transactions )
SELECT fullvisitorid, DATE_DIFF(PARSE_DATE('%Y%m%d',date_event), PARSE_DATE('%Y%m%d', date_first_visit),DAY) AS time,
transaction, device FROM mortality_table"""
query_results = client.query(query) ; query_results = query_results.result()
#Résutats de la reqête
fullvisitorid = [] ; time = [] ; device = [] ; transaction = []
for row in query_results:
fullvisitorid.append(row[0])
time.append(row[1])
transaction.append(row[2])
device.append(row[3])
BigQuery_table_2 = {"fullvisitorid":fullvisitorid, "time":time, "transaction":transaction, "device":device}
BigQuery_table_2 = pd.DataFrame(BigQuery_table_2)
survival_data_2 = pd.DataFrame(np.c_[BigQuery_table_2.iloc[:,1:4]],
columns = col, index = BigQuery_table_2['fullvisitorid'])
#On crée le modèle
kmf_2 = KaplanMeierFitter() ; kmf_2.fit(survival_data_2['time'],survival_data_2['transaction'])
#Courbe de survie
plt.figure(figsize=(15,10)) ; plt.ylim([0,35,1]) ; kmf.plot(label='conversion') ; kmf_2.plot(label='retention')
plt.xlabel('time in days') ; plt.ylabel('survival probability') ; plt.title("Kaplan Meier Survival Functionsn")
#On divise le jeu de données en fonction des segment
desktop_2 = survival_data_2.query("device == 'desktop'")
mobile_2 = survival_data_2.query("device == 'mobile'")
tablet_2 = survival_data_2.query("device == 'tablet'")
#On crée le modèle
kmf_desktop_2 = KaplanMeierFitter() ; kmf_mobile_2 = KaplanMeierFitter() ; kmf_tablet_2 = KaplanMeierFitter()
kmf_desktop_2.fit(desktop_2['time'],desktop_2['transaction'])
kmf_mobile_2.fit(mobile_2['time'],mobile_2['transaction'])
kmf_tablet_2.fit(tablet_2['time'],tablet_2['transaction'])
#Courbes de survie en fonction des segments
plt.figure(figsize=(15,10)) ; kmf_desktop_2.plot(label='desktop') ; kmf_mobile_2.plot(label='mobile')
kmf_tablet_2.plot(label='tablet') ; plt.xlabel("time in days") ; plt.ylabel("survival probability")
plt.title("Kaplan Meier Survival Functions") ; plt.ylim([0.35,1])