import pandas as pd import numpy as np import seaborn as sn from sklearn.model_selection import cross_val_score from sklearn import ensemble from sklearn.preprocessing import LabelEncoder import re %matplotlib inline
action_train=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\trainingset\action_train.csv') orderFuture_train=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\trainingset\orderFuture_train.csv') orderHistory_train=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\trainingset\orderHistory_train.csv') userComment_train=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\trainingset\userComment_train.csv') userProfile_train=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\trainingset\userProfile_train.csv') action_train2=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\trainingset\action_train2.csv')
action_test=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\test\action_test.csv') orderFuture_test=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\test\orderFuture_test.csv') orderHistory_test=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\test\orderHistory_test.csv') userComment_test=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\test\userComment_test.csv') userProfile_test=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\test\userProfile_test.csv') action_test2=pd.read_csv(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\data\test\action_test2.csv')
##########对训练集构建特征工程############# ####构建新特征变量orderType1，曾经购买过精品服务的设为1，其余设为0 orderHistory_train1=orderHistory_train.drop_duplicates(['userid']) train_ordertype_number=[] for i in range(len(orderHistory_train1.index)): count_number=list(orderHistory_train['userid']).count(orderHistory_train.loc[orderHistory_train1.index[i],'userid']) count_1_number=list(orderHistory_train.loc[orderHistory_train1.index[i]:orderHistory_train1.index[i]+count_number-1,'orderType']).count(1) train_ordertype_number.append(count_1_number) orderHistory_train1['orderType1']=train_ordertype_number
####构建新特征变量，每个用户使用app的次数operation_number和总时长operation_time和1到9的点击次数 action_train1=action_train.drop_duplicates('userid') number=[] time_longth=[] for i in range(len(action_train1.index)-1): count_number=action_train1.index[i+1]-action_train1.index[i] number.append(count_number) time=action_train.loc[action_train1.index[i+1]-1,'actionTime']-action_train.loc[action_train1.index[i],'actionTime'] time_longth.append(time) number.append(len(action_train.index)-action_train1.index[-1]) time_longth.append(action_train.loc[action_train.index[-1],'actionTime']-action_train.loc[action_train1.index[-1],'actionTime']) action_train1['operation_number']=number action_train1['operation_time']=time_longth action_train1.drop(['actionType','actionTime'],axis=1,inplace=True)
action_train['actionType']=action_train['actionType'].astype('object') action_train2=pd.get_dummies(action_train) action_train3=action_train.drop_duplicates(['userid']) train_actionType_1_number=[] train_actionType_2_number=[] train_actionType_3_number=[] train_actionType_4_number=[] train_actionType_5_number=[] train_actionType_6_number=[] train_actionType_7_number=[] train_actionType_8_number=[] train_actionType_9_number=[] for i in range(len(action_train3.index)): each_actionType_1=action_train2[action_train2['userid']==action_train2.loc[action_train3.index[i],'userid']]['actionType_1'].sum() each_actionType_2=action_train2[action_train2['userid']==action_train2.loc[action_train3.index[i],'userid']]['actionType_2'].sum() each_actionType_3=action_train2[action_train2['userid']==action_train2.loc[action_train3.index[i],'userid']]['actionType_3'].sum() each_actionType_4=action_train2[action_train2['userid']==action_train2.loc[action_train3.index[i],'userid']]['actionType_4'].sum() each_actionType_5=action_train2[action_train2['userid']==action_train2.loc[action_train3.index[i],'userid']]['actionType_5'].sum() each_actionType_6=action_train2[action_train2['userid']==action_train2.loc[action_train3.index[i],'userid']]['actionType_6'].sum() each_actionType_7=action_train2[action_train2['userid']==action_train2.loc[action_train3.index[i],'userid']]['actionType_7'].sum() each_actionType_8=action_train2[action_train2['userid']==action_train2.loc[action_train3.index[i],'userid']]['actionType_8'].sum() each_actionType_9=action_train2[action_train2['userid']==action_train2.loc[action_train3.index[i],'userid']]['actionType_9'].sum() each_sum=each_actionType_1+each_actionType_2+each_actionType_3+each_actionType_4+each_actionType_5+each_actionType_6+each_actionType_7+each_actionType_8+each_actionType_9 train_actionType_1_number.append(each_actionType_1/each_sum) train_actionType_2_number.append(each_actionType_2/each_sum) train_actionType_3_number.append(each_actionType_3/each_sum) train_actionType_4_number.append(each_actionType_4/each_sum) train_actionType_5_number.append(each_actionType_5/each_sum) train_actionType_6_number.append(each_actionType_6/each_sum) train_actionType_7_number.append(each_actionType_7/each_sum) train_actionType_8_number.append(each_actionType_8/each_sum) train_actionType_9_number.append(each_actionType_9/each_sum) action_train3['orderType_1']=train_actionType_1_number action_train3['orderType_2']=train_actionType_2_number action_train3['orderType_3']=train_actionType_3_number action_train3['orderType_4']=train_actionType_4_number action_train3['orderType_5']=train_actionType_5_number action_train3['orderType_6']=train_actionType_6_number action_train3['orderType_7']=train_actionType_7_number action_train3['orderType_8']=train_actionType_8_number action_train3['orderType_9']=train_actionType_9_number action_train3.drop(['actionType','actionTime'],axis=1,inplace=True)
train1=pd.merge(userProfile_train,userComment_train,how='outer',on=['userid']) train2=pd.merge(train1,orderHistory_train1,how='outer',on=['userid']) train2.drop(['orderid_x','tags','commentsKeyWords','orderid_y','orderTime','city','country','continent','orderType'],axis=1,inplace=True) train3=pd.merge(train2,orderFuture_train,how='outer',on=['userid']) train3=pd.merge(train3,action_train1,how='outer',on=['userid']) train3=pd.merge(train3,action_train3,how='outer',on=['userid']) train3.info()
#对测试集构建特征工程 orderHistory_test1=orderHistory_test.drop_duplicates(['userid']) test_ordertype_number=[] for i in range(len(orderHistory_test1.index)): count_number=list(orderHistory_test['userid']).count(orderHistory_test.loc[orderHistory_test1.index[i],'userid']) count_1_number=list(orderHistory_test.loc[orderHistory_test1.index[i]:orderHistory_test1.index[i]+count_number-1,'orderType']).count(1) test_ordertype_number.append(count_1_number) orderHistory_test1['orderType1']=test_ordertype_number
action_test1=action_test.drop_duplicates('userid') number=[] time_longth=[] for i in range(len(action_test1.index)-1): count_number=action_test1.index[i+1]-action_test1.index[i] number.append(count_number) time=action_test.loc[action_test1.index[i+1]-1,'actionTime']-action_test.loc[action_test1.index[i],'actionTime'] time_longth.append(time) number.append(len(action_test.index)-action_test1.index[-1]) time_longth.append(action_test.loc[action_test.index[-1],'actionTime']-action_test.loc[action_test1.index[-1],'actionTime']) action_test1['operation_number']=number action_test1['operation_time']=time_longth action_test1.drop(['actionType','actionTime'],axis=1,inplace=True)
action_test['actionType']=action_test['actionType'].astype('object') action_test2=pd.get_dummies(action_test) action_test3=action_test.drop_duplicates(['userid']) test_actionType_1_number=[] test_actionType_2_number=[] test_actionType_3_number=[] test_actionType_4_number=[] test_actionType_5_number=[] test_actionType_6_number=[] test_actionType_7_number=[] test_actionType_8_number=[] test_actionType_9_number=[] for i in range(len(action_test3.index)): each_actionType_1=action_test2[action_test2['userid']==action_test2.loc[action_test3.index[i],'userid']]['actionType_1'].sum() each_actionType_2=action_test2[action_test2['userid']==action_test2.loc[action_test3.index[i],'userid']]['actionType_2'].sum() each_actionType_3=action_test2[action_test2['userid']==action_test2.loc[action_test3.index[i],'userid']]['actionType_3'].sum() each_actionType_4=action_test2[action_test2['userid']==action_test2.loc[action_test3.index[i],'userid']]['actionType_4'].sum() each_actionType_5=action_test2[action_test2['userid']==action_test2.loc[action_test3.index[i],'userid']]['actionType_5'].sum() each_actionType_6=action_test2[action_test2['userid']==action_test2.loc[action_test3.index[i],'userid']]['actionType_6'].sum() each_actionType_7=action_test2[action_test2['userid']==action_test2.loc[action_test3.index[i],'userid']]['actionType_7'].sum() each_actionType_8=action_test2[action_test2['userid']==action_test2.loc[action_test3.index[i],'userid']]['actionType_8'].sum() each_actionType_9=action_test2[action_test2['userid']==action_test2.loc[action_test3.index[i],'userid']]['actionType_9'].sum() each_sum=each_actionType_1+each_actionType_2+each_actionType_3+each_actionType_4+each_actionType_5+each_actionType_6+each_actionType_7+each_actionType_8+each_actionType_9 test_actionType_1_number.append(each_actionType_1/each_sum) test_actionType_2_number.append(each_actionType_2/each_sum) test_actionType_3_number.append(each_actionType_3/each_sum) test_actionType_4_number.append(each_actionType_4/each_sum) test_actionType_5_number.append(each_actionType_5/each_sum) test_actionType_6_number.append(each_actionType_6/each_sum) test_actionType_7_number.append(each_actionType_7/each_sum) test_actionType_8_number.append(each_actionType_8/each_sum) test_actionType_9_number.append(each_actionType_9/each_sum) action_test3['orderType_1']=test_actionType_1_number action_test3['orderType_2']=test_actionType_2_number action_test3['orderType_3']=test_actionType_3_number action_test3['orderType_4']=test_actionType_4_number action_test3['orderType_5']=test_actionType_5_number action_test3['orderType_6']=test_actionType_6_number action_test3['orderType_7']=test_actionType_7_number action_test3['orderType_8']=test_actionType_8_number action_test3['orderType_9']=test_actionType_9_number action_test3.drop(['actionType','actionTime'],axis=1,inplace=True)
test1=pd.merge(userProfile_test,userComment_test,how='outer',on=['userid']) test2=pd.merge(test1,orderHistory_test1,how='outer',on=['userid']) test2.drop(['orderid_x','tags','commentsKeyWords','orderid_y','orderTime','city','country','continent','orderType'],axis=1,inplace=True) test3=pd.merge(test2,orderFuture_test,how='outer',on=['userid']) test3=pd.merge(test3,action_test1,how='outer',on=['userid']) test3=pd.merge(test3,action_test3,how='outer',on=['userid']) test3.info()
#########数据清洗######### train3['rating'].fillna(train3['rating'].mode()[0],inplace=True) train3['orderType1'].fillna(train3['orderType1'].mode()[0],inplace=True) train3.drop(['userid','gender','province','age'],axis=1,inplace=True) test3['rating'].fillna(test3['rating'].mode()[0],inplace=True) test3['orderType1'].fillna(test3['orderType1'].mode()[0],inplace=True) test3.drop(['userid','gender','province','age'],axis=1,inplace=True)
###RandomForest###### rf=ensemble.RandomForestClassifier(50) #features=['rating','orderType1','operation_number'] features=list(train3.columns) features.remove('orderType') X=train3[features] print(X) y=train3['orderType'] model=rf.fit(X,y) scores=cross_val_score(rf,X,y,cv=5,scoring='accuracy') print(np.mean(scores))
X=test3[features] y=model.predict(X) np.savetxt('E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\submission3.csv',np.c_[orderFuture_test['userid'],y],delimiter=',',header='userid,orderType',comments='',fmt='%d')
########XGBoost######## import xgboost as xgb import time from sklearn.model_selection import train_test_split #feature engineering
train=train3 train_xy,val=train_test_split(train,test_size=0.2,random_state=1) y=train_xy['orderType'] X=train_xy.drop(['orderType'],axis=1) val_y=val['orderType'] val_x=val.drop(['orderType'],axis=1) xgb_train=xgb.DMatrix(X,label=y) xgb_val=xgb.DMatrix(val_x,label=val_y) xgb_test=xgb.DMatrix(test3)
#设置模型参数，建立XGB模型 start_time=time.time()
params={
'booster':'gbtree',
'objective': 'multi:softmax', #多分类的问题
'num_class':2, # 类别数，与 multisoftmax 并用
'gamma':0.1, # 用于控制是否后剪枝的参数,越大越保守，一般0.1、0.2这样子。
'max_depth':12, # 构建树的深度，越大越容易过拟合
'lambda':2, # 控制模型复杂度的权重值的L2正则化项参数，参数越大，模型越不容易过拟合。
'subsample':0.7, # 随机采样训练样本
'colsample_bytree':0.7, # 生成树时进行的列采样
'min_child_weight':3,
'silent':0 ,#设置成1则没有运行信息输出，最好是设置为0.
'eta': 0.007, # 如同学习率
'seed':1000,
'nthread':2,# cpu 线程数
#'eval_metric': 'auc'
}
plst=list(params.items()) number_rounds=1000 watch_list=[(xgb_train,'train'),(xgb_val,'val')] model=xgb.train(plst,xgb_train,number_rounds,watch_list,early_stopping_rounds=100) model.save_model(r'E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\xgb.model')
#预测 print("best best_ntree_limit",model.best_ntree_limit )
print ("跑到这里了model.predict") preds = model.predict(xgb_test,ntree_limit=model.best_ntree_limit)
np.savetxt('E:\Program_Files\datacastal\practice\huangbaoche\huangbaoche_match\submission4.csv',np.c_[orderFuture_test['userid'],preds],delimiter=',',header='userid,orderType',comments='',fmt='%d')
#输出运行时长 cost_time = time.time()-start_time print ("xgboost success!",'\n',"cost time:",cost_time,"(s)")