gbc.py

from sklearn.ensemble import GradientBoostingClassifier
from sklearn.metrics import roc_curve,roc_auc_score
from sklearn.utils import shuffle
from sklearn.externals import joblib
from matplotlib import pyplot
import pickle
import numpy as np
from os import path

with open('train.pkl','rb') as f:
    product_train=pickle.load(f)
    retrieved_train=pickle.load(f)
    user_train=pickle.load(f)
    labels_train=pickle.load(f)

with open('test.pkl','rb') as f:
    product_test=pickle.load(f)
    retrieved_test=pickle.load(f)
    user_test=pickle.load(f)
    labels_test=pickle.load(f)

with open('history.pkl','rb') as f:
    history=pickle.load(f)

train_data=[]
test_data=[]

for i,prod_len in enumerate(product_train):
    if prod_len>49:
        prod_len=50
        zero_padding_len=0
    else:
        zero_padding_len=50-prod_len
    temp=[user_train[i],retrieved_train[i]]
    temp+=[0]*zero_padding_len
    # add 50(at most) recent history 
    temp_history=history[user_train[i]][-prod_len:]
    temp+=temp_history
    assert len(temp)==52
    
    train_data.append(temp)

for i,prod_len in enumerate(product_test):
    if prod_len>49:
        prod_len=50
        zero_padding_len=0
    else:
        zero_padding_len=50-prod_len
    temp=[user_test[i],retrieved_test[i]]
    temp+=[0]*zero_padding_len
    temp_history=history[user_test[i]][-prod_len:]
    temp+=temp_history

    assert len(temp)==52
    
    test_data.append(temp)

train_data,labels_train=shuffle(train_data,labels_train,random_state=2000)

if path.exists('gbc.pkl'):
    gbc=joblib.load('gbc.pkl')
else:
    gbc=GradientBoostingClassifier()
    gbc.fit(train_data,labels_train)
    joblib.dump(gbc,'gbc.pkl')

print(gbc.feature_importances_)

score=gbc.score(test_data,labels_test)

print('The test score is : {:.4f}'.format(score))

prob= gbc.predict_proba(test_data)[:,1]

fpr,tpr,threshold=roc_curve(labels_test,prob,pos_label=1)

pyplot.plot(fpr,tpr)
pyplot.xlabel('False positive rate')
pyplot.ylabel('True positive rate')
pyplot.title('ROC curve')
pyplot.legend(loc='best')

auc_score=roc_auc_score(labels_test,prob)

print('auc score: {:.4f}'.format(auc_score))

pyplot.show()