阿里AI天池大赛-天猫复购预测-基于XGBoost模型预测

1：报名地址

https://tianchi.aliyun.com/competition/entrance/231576/score

2：排名分数 

3：模型源码

 废话不多说，直接上源码

import gc
import pandas as pd
# 用户行为，使用format1进行加载

# 加载全量样本
user_log = pd.read_csv('./data_format1/user_log_format1.csv', dtype={'time_stamp':'str'})
user_info = pd.read_csv('./data_format1/user_info_format1.csv')
train_data1 = pd.read_csv('./data_format1/train_format1.csv')
submission = pd.read_csv('./data_format1/test_format1.csv')
train_data = pd.read_csv('./data_format2/train_format2.csv')


train_data1['origin'] = 'train'
submission['origin'] = 'test'
matrix = pd.concat([train_data1, submission], ignore_index=True, sort=False)
#print(matrix)

matrix.drop(['prob'], axis=1, inplace=True)
# 连接user_info表，通过user_id关联
matrix = matrix.merge(user_info, on='user_id', how='left')
# 使用merchant_id（原列名seller_id）
user_log.rename(columns={'seller_id':'merchant_id'}, inplace=True)
# 格式化
user_log['user_id'] = user_log['user_id'].astype('int32')
user_log['merchant_id'] = user_log['merchant_id'].astype('int32')
user_log['item_id'] = user_log['item_id'].astype('int32')
user_log['cat_id'] = user_log['cat_id'].astype('int32')
user_log['brand_id'].fillna(0, inplace=True)
user_log['brand_id'] = user_log['brand_id'].astype('int32')
user_log['time_stamp'] = pd.to_datetime(user_log['time_stamp'], format='%H%M')
# 1 for <18; 2 for [18,24]; 3 for [25,29]; 4 for [30,34]; 5 for [35,39]; 6 for [40,49]; 7 and 8 for >= 50; 0 and NULL for unknown
matrix['age_range'].fillna(0, inplace=True)
# 0:female, 1:male, 2:unknown
matrix['gender'].fillna(2, inplace=True)
matrix['age_range'] = matrix['age_range'].astype('int8')
matrix['gender'] = matrix['gender'].astype('int8')
matrix['label'] = matrix['label'].astype('str')
matrix['user_id'] = matrix['user_id'].astype('int32')
matrix['merchant_id'] = matrix['merchant_id'].astype('int32')
del user_info, train_data1
gc.collect()
print(matrix)

# User特征处理
groups = user_log.groupby(['user_id'])
# 用户交互行为数量 u1
temp = groups.size().reset_index().rename(columns={0:'u1'})
matrix = matrix.merge(temp, on='user_id', how='left')
# 使用agg 基于列的聚合操作，统计唯一值的个数 item_id, cat_id, merchant_id, brand_id
#temp = groups['item_id', 'cat_id', 'merchant_id', 'brand_id'].nunique().reset_index().rename(columns={'item_id':'u2', 'cat_id':'u3', 'merchant_id':'u4', 'brand_id':'u5'})
# 对于每个user_id 不重复的item_id的数量 => u2
temp = groups['item_id'].agg([('u2', 'nunique')]).reset_index()
matrix = matrix.merge(temp, on='user_id', how='left')
# 对于每个user_id 不重复的cat_id的数量 => u3
temp = groups['cat_id'].agg([('u3', 'nunique')]).reset_index()
matrix = matrix.merge(temp, on='user_id', how='left')
temp = groups['merchant_id'].agg([('u4', 'nunique')]).reset_index()
matrix = matrix.merge(temp, on='user_id', how='left')
temp = groups['brand_id'].agg([('u5', 'nunique')]).reset_index()
matrix = matrix.merge(temp, on='user_id', how='left')

# 时间间隔特征 u6 按照小时
# 对于每个user_id 计算time_stamp的最小时间 => F_time, 最大时间max => L_time
temp = groups['time_stamp'].agg([('F_time', 'min'), ('L_time', 'max')]).reset_index()
temp['u6'] = (temp['L_time'] - temp['F_time']).dt.seconds/3600
matrix = matrix.merge(temp[['user_id', 'u6']], on='user_id', how='left')
# 统计操作类型为0，1，2，3的个数
temp = groups['action_type'].value_counts().unstack().reset_index().rename(columns={0:'u7', 1:'u8', 2:'u9', 3:'u10'})
matrix = matrix.merge(temp, on='user_id', how='left')
#print(matrix)

# 商家特征处理
groups = user_log.groupby(['merchant_id'])
# 商家被交互行为数量 m1
temp = groups.size().reset_index().rename(columns={0:'m1'})
matrix = matrix.merge(temp, on='merchant_id', how='left')
# 统计商家被交互的user_id, item_id, cat_id, brand_id 唯一值
temp = groups['user_id', 'item_id', 'cat_id', 'brand_id'].nunique().reset_index().rename(columns={'user_id':'m2', 'item_id':'m3', 'cat_id':'m4', 'brand_id':'m5'})
matrix = matrix.merge(temp, on='merchant_id', how='left')
# 统计商家被交互的action_type 唯一值
temp = groups['action_type'].value_counts().unstack().reset_index().rename(columns={0:'m6', 1:'m7', 2:'m8', 3:'m9'})
matrix = matrix.merge(temp, on='merchant_id', how='left')
# 按照merchant_id 统计随机负采样的个数
temp = train_data[train_data['label']==-1].groupby(['merchant_id']).size().reset_index().rename(columns={0:'m10'})
matrix = matrix.merge(temp, on='merchant_id', how='left')
#print(matrix)

# 按照user_id, merchant_id分组
groups = user_log.groupby(['user_id', 'merchant_id'])
temp = groups.size().reset_index().rename(columns={0:'um1'}) #统计行为个数
matrix = matrix.merge(temp, on=['user_id', 'merchant_id'], how='left')
temp = groups['item_id', 'cat_id', 'brand_id'].nunique().reset_index().rename(columns={'item_id':'um2', 'cat_id':'um3', 'brand_id':'um4'}) #统计item_id, cat_id, brand_id唯一个数
matrix = matrix.merge(temp, on=['user_id', 'merchant_id'], how='left')
temp = groups['action_type'].value_counts().unstack().reset_index().rename(columns={0:'um5', 1:'um6', 2:'um7', 3:'um8'})#统计不同action_type唯一个数
matrix = matrix.merge(temp, on=['user_id', 'merchant_id'], how='left')
temp = groups['time_stamp'].agg([('first', 'min'), ('last', 'max')]).reset_index()
temp['um9'] = (temp['last'] - temp['first']).dt.seconds/3600
temp.drop(['first', 'last'], axis=1, inplace=True)
matrix = matrix.merge(temp, on=['user_id', 'merchant_id'], how='left') #统计时间间隔
#print(matrix)

#用户购买点击比
matrix['r1'] = matrix['u9']/matrix['u7'] 
#商家购买点击比
matrix['r2'] = matrix['m8']/matrix['m6'] 
#不同用户不同商家购买点击比
matrix['r3'] = matrix['um7']/matrix['um5']
matrix.fillna(0, inplace=True)
# # 修改age_range字段名称为 age_0, age_1, age_2... age_8
temp = pd.get_dummies(matrix['age_range'], prefix='age')
matrix = pd.concat([matrix, temp], axis=1)
temp = pd.get_dummies(matrix['gender'], prefix='g')
matrix = pd.concat([matrix, temp], axis=1)
matrix.drop(['age_range', 'gender'], axis=1, inplace=True)
print(matrix)

# 分割训练数据和测试数据
train_data = matrix[matrix['origin'] == 'train'].drop(['origin'], axis=1)
test_data = matrix[matrix['origin'] == 'test'].drop(['label', 'origin'], axis=1)
train_X, train_y = train_data.drop(['label'], axis=1), train_data['label']
del temp, matrix
gc.collect()

# 使用机器学习工具
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.linear_model import LinearRegression
from sklearn.metrics import classification_report
import xgboost as xgb
# 将训练集进行切分，20%用于验证
X_train, X_valid, y_train, y_valid = train_test_split(train_X, train_y, test_size=.2)

# 使用XGBoost
model = xgb.XGBClassifier(
    max_depth=8,
    n_estimators=2000,
    min_child_weight=300, 
    colsample_bytree=0.8, 
    subsample=0.8, 
    eta=0.3,    
    seed=42    
)
model.fit(
    X_train, y_train,
    eval_metric='auc', eval_set=[(X_train, y_train), (X_valid, y_valid)],
    verbose=True,
    #早停法，如果auc在10epoch没有进步就stop
    early_stopping_rounds=30 
)

model.fit(X_train, y_train)

prob = model.predict_proba(test_data)
submission['prob'] = pd.Series(prob[:,1])
submission.drop(['origin'], axis=1, inplace=True)
submission.to_csv('BaseLine_0.676357.csv', index=False)

 4：提分要领

       用户特征处理

        时间间隔特征

        商家特征处理

        购买点击比

5：相关知识补充

1：DIN模型

2：DIEN模型

 3：DSIN模型