/
main.py
106 lines (91 loc) · 5.01 KB
/
main.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
import pandas as pd
import numpy as np
from sklearn.linear_model import LogisticRegression
from sklearn import metrics
from sklearn.metrics import roc_curve, auc, confusion_matrix
from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier, GradientBoostingClassifier, GradientBoostingRegressor
from sklearn.ensemble.partial_dependence import plot_partial_dependence, partial_dependence
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import GridSearchCV, cross_val_predict
import matplotlib.pyplot as plt
from feature_engineering import build_X, build_y
from models import logisticModel, randomForestModel, GradientBoost, adaBoost, GBC_Logistic, RF_Logistic
from costbenefit_analysis import standard_confusion_matrix, profit_curve, plot_model_profits
def get_data(n_days):
y = build_y(churnTrainDF, delta_days=n_days)
X = build_X(churnTrainDF, delta_days=n_days)
y_test = build_y(churnTestDF, delta_days=n_days)
X_test = build_X(churnTestDF, delta_days=n_days)
return y, X, y_test, X_test
if __name__ == '__main__':
print("$ Reading Data")
churnTrainDF = pd.read_csv('data/churn_train.csv',parse_dates=['last_trip_date','signup_date'])
churnTestDF = pd.read_csv('data/churn_test.csv',parse_dates=['last_trip_date','signup_date'])
churn_definitions = ['30 days', '90 days']
for n_days in churn_definitions:
y, X, y_test, X_test = get_data(n_days)
print("$ Modeling")
predictedLogistic = logisticModel(X, y, X_test)
predictedRF = randomForestModel(X, y, X_test)
predictedGradient = GradientBoost(X,y, X_test)
predictedAda = adaBoost(X, y, X_test)
predicted_G_L = GBC_Logistic(X,y,X_test)
predicted_RF_L = RF_Logistic(X,y,X_test)
# Plot the ROC curves for different models
print("$ ROC Curves")
fpr_l, tpr_l, thresholds_l = roc_curve(y_test, predictedLogistic)
auc_score_l = auc(fpr_l, tpr_l)
fpr_rf, tpr_rf, thresholds_rf = roc_curve(y_test, predictedRF)
auc_score_rf = auc(fpr_rf, tpr_rf)
fpr_a, tpr_a, thresholds_a = roc_curve(y_test, predictedAda)
auc_score_a = auc(fpr_a, tpr_a)
fpr_g, tpr_g, thresholds_g = roc_curve(y_test, predictedGradient)
auc_score_g = auc(fpr_g, tpr_g)
fpr_g_l, tpr_g_l, thresholds_g_l = roc_curve(y_test, predicted_G_L)
auc_score_g_l = auc(fpr_g_l, tpr_g_l)
fpr_rf_l, tpr_rf_l, thresholds_rf_l = roc_curve(y_test, predicted_RF_L)
auc_score_rf_l = auc(fpr_rf_l, tpr_rf_l)
plt.figure(1)
plt.plot([0, 1], [0, 1], 'k--')
plt.plot(fpr_l, tpr_l, label= ('Logistic, AUC: {0:.2f}'.format(auc_score_l)))
plt.plot(fpr_rf, tpr_rf, label=('Random Forest, AUC: {0:.2f}'.format(auc_score_rf)))
plt.plot(fpr_a, tpr_a, label=('AdaBoosting, AUC: {0:.2f}'.format(auc_score_a)))
plt.plot(fpr_g, tpr_g, label=('GradientBoosting, AUC: {0:.2f}'.format(auc_score_g)))
plt.plot(fpr_g_l, tpr_g_l, label=('GBM + LG, AUC: {0:.2f}'.format(auc_score_g_l)))
plt.plot(fpr_rf_l, tpr_rf_l, label=('RF + LG, AUC: {0:.2f}'.format(auc_score_rf_l)))
plt.xlabel('False positive rate')
plt.ylabel('True positive rate')
plt.title('{} ROC curve'.format(n_days))
plt.legend(loc='best')
plt.savefig('{} roc.png'.format(n_days))
plt.close()
# Plot PDPs
print("$ PDP Plots")
clf = GradientBoostingClassifier(learning_rate=0.5, n_estimators=200, max_depth=3)
fit = clf.fit(X, y)
names = ['avg_dist_log', 'avg_rating_by_driver_log', 'avg_rating_of_driver_log',
'avg_surge', 'surge_pct', 'trips_in_first_30_days', 'weekday_pct',
"city_King's Landing", 'city_Winterfell', 'phone_iPhone',
'luxury_car_user_True', 'user_lifespan', 'user_rated_driver',
'user_rated_driver_avg_rating_of_driver', 'city_Astapor', 'phone_Android',
'luxury_car_user_False', 'avg_dist', 'avg_rating_by_driver', 'avg_rating_of_driver']
features = [5,9,15,3,4,6,7,8,14,16,10,11,17,18,19]
pdp = plt.figure(figsize=(16,8))
# plt.title('{} Partial Dependency Plot'.format(n_days))
fig, axs = plot_partial_dependence(fit, X, features, feature_names=names, n_jobs=2, grid_resolution=50)
fig.set_size_inches(18,12)
fig.tight_layout()
fig.savefig('{} pdp.png'.format(n_days))
plt.close(pdp)
# plot profit curves
cost_benefit = np.array([[20, -20], [0, 0]])
profits = profit_curve(cost_benefit,predictedGradient,y_test)
max_profit = max(profits)[0]
max_threshold = max(profits)[1]
print(max_profit, max_threshold)
y_predict = predictedGradient >= max_threshold
confusion_matrix = standard_confusion_matrix(y_test,y_predict)
print('confusion matrix',confusion_matrix)
threshold_profit = np.sum(confusion_matrix * cost_benefit) / y_test.shape[0]
print('profit',threshold_profit)
plot_model_profits(profits,save_path='{} profit.png'.format(n_days),n_days=n_days)