天池新人赛-天池新人实战赛o2o优惠券使用预测（…

1 import numpy as np
2 import pandas as pd
3 
4 #导入数据
5 train_online = pd.read_csv('ccf_online_stage1_train.csv')
6 train_offline = pd.read_csv('ccf_offline_stage1_train.csv')
7 test = pd.read_csv('ccf_offline_stage1_test_revised.csv')

1 #将数据合并，以便统一对数据进行处理。都是线下数据
2 all_offline = pd.concat([train_offline,test])

1 #查看每一列的异常值
2 f = lambda x:sum(x.isnull())
3 all_offline.apply(f)

1 #Coupon_id 优惠券ID的具体数值意义不大，因此我们把他转换成：是否有优惠券
2 def has_coup(x):
3     if x['Coupon_id'] != 'null':
4         return 1
5     else:
6         return 0
7     
8 all_offline['has_coup']=all_offline.apply(has_coup,axis=1)

 1 #在进一步想，优惠力度会影响优惠券使用的概率，x:y这种满减的类型，x具体是多少，势必也会影响优惠券使用率
 2 #讲满x元的标出x元
 3 def discount_limit(y):
 4     if y['Discount_rate'] == 'null' and y['Date_received'] == 'null':
 5         return 'null'
 6     elif re.match(regex,y['Discount_rate']):
 7         num_min,num_max=y['Discount_rate'].split(':')
 8         return num_min
 9     else:
10         return 0
11 
12 all_offline['discount_limit'] = all_offline.apply(discount_limit,axis=1)
13 all_offline.head(10)

 1 #由于赛题需要的是，优惠券领取后15天的使用概率
 2 #因此，我们在is_used的基础上，在对领券时间 Date_received 和使用时间Date，进行比较，判断是否在15天内使用
 3 #时间比较
 4 import datetime
 5 #标注15天内使用优惠券的情况
 6 def used_in_15days(z):
 7     if z['is_used'] == 1 and z['Date'] != 'null' and z['Date_received'] != 'null':
 8         days= (datetime.datetime.strptime(z['Date'],"%Y%m%d")-datetime.datetime.strptime(z['Date_received'],"%Y%m%d"))
 9         if days.days < 15:
10             return 1
11         else:
12             return 0
13     else:
14         return 0
15         
16 all_offline['used_in_15days']=all_offline.apply(used_in_15days,axis=1)

1 #再来观察discount_percent，discount_limit这2个特征，看数据的分布情况。
2 all_offline['discount_percent'].value_counts()

1 all_offline['discount_limit'].value_counts()

#将discount_percent分段
def discount_percent_layer(columns):
    if columns['discount_percent']=='null':
        return 'null'
    
    columns['discount_percent']=float(columns['discount_percent'])
    if columns['discount_percent'] <= 0.1:
        return 0.1
    elif columns['discount_percent'] <= 0.2:
        return 0.2
    elif columns['discount_percent'] <= 0.3:
        return 0.3
    elif columns['discount_percent'] <= 0.4:
        return 0.4
    else:
        return 0.5

all_offline['discount_percent_layer']=all_offline.apply(discount_percent_layer,axis=1)
all_offline['discount_percent_layer'].value_counts()

 1 #将discount_limit分段
 2 def discount_limit_layer(columns):
 3     if columns =='null':
 4         return 'null'
 5     
 6     columns=int(columns)
 7     if columns <= 10:
 8         return 10
 9     elif columns <= 20:
10         return 20
11     elif columns <= 30:
12         return 30
13     elif columns <= 50:
14         return 50
15     elif columns <= 100:
16         return 100
17     elif columns <= 200:
18         return 200
19     else:
20         return 300
21 
22 all_offline['discount_limit_layer']=all_offline['discount_limit'].apply(discount_limit_layer)
23 all_offline['discount_limit_layer'].value_counts()

1 #剩下Distance，看下Distance的分布情况
2 all_offline['Distance'].value_counts()

1 #保存数据，以便后期使用起来方便
2 train_finall,test_finall = all_offline[:train_offline.shape[0]],all_offline[train_offline.shape[0]:]
3 all_offline.to_csv(r'output\all_offline.csv')
4 train_finall.to_csv(r'output\train_finall.csv')
5 test_finall.to_csv(r'output\test_finall.csv')

1 #one_hot处理
2 all_offline_new=all_offline.drop(
3         ['Coupon_id','Date','Date_received','Discount_rate','Merchant_id',
4          'User_id','discount_percent','discount_limit'],axis=1)
5 all_offline_new=pd.get_dummies(all_offline_new)

 1 #把测试集跟验证集分开
 2 train01,test01=all_offline_new[:len(train_offline)],all_offline_new[len(train_offline):]
 3 
 4 #把没有领券的去掉
 5 train02=train01[train01['has_coup']==1]
 6 
 7 #由于特征集 都是领券的人，故把 has_coup 字段删掉
 8 train02=train02.drop(['has_coup'],axis=1)
 9 test01=test01.drop(['has_coup'],axis=1)
10 
11 x_train=train02.drop(['used_in_15days'],axis=1)
12 y_train=pd.DataFrame({"used_in_15days":train02['used_in_15days']})
13 x_text=test01.drop(['used_in_15days'],axis=1)

 1 #建模
 2 from sklearn.linear_model import LinearRegression
 3 
 4 clf=LinearRegression()
 5 clf.fit(x_train,y_train)
 6 
 7 #用模型进行预测
 8 predict=clf.predict(x_text)
 9 
10 
11 result=pd.read_csv('ccf_offline_stage1_test_revised.csv')
12 result['probability']=predict
13 
14 result=result.drop(['Merchant_id','Discount_rate','Distance'],axis=1)
15 
16 
17 #发现最终预测有负值，直接归为0
18 result['probability']=result['probability'].apply(lambda x: 0 if x<0 else x)
19 
20 result.to_csv(r'output/sample_submission.csv',index=False)