2021 CCF基于UEBA的用户上网异常行为分析baseline线上0.90

2021CCF BDCI

今年CCF又来了，每年都有大佬选手夺冠，也有黑马新人突出重围，对于新人来说一份baseline是很好的起点，可以更快入门数据竞赛。(大佬请忽略！！！)

基于UEBA的用户上网异常行为分析

结构化数据比较好入手，由于贷款违约数据有问题，所以选择了另外一个结构化赛题写了一个baseline，我只提交了一次，线上有0.8994，虽然比不上前排大佬的分数，但是对于入门来说还是可以参考下，而且提升空间还很大！

比赛地址链接：https://www.datafountain.cn/competitions/520

数据列表

话不多说，直接上代码：

import pandas as pd
import numpy as np
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import KFold
from sklearn.metrics import mean_squared_error
import lightgbm as lgb
import xgboost as xgb
from sklearn.model_selection import KFold
from sklearn.preprocessing import LabelEncoder
from sklearn.metrics import mean_squared_error
import math
from sklearn import ensemble
from datetime import datetime
df_train = pd.read_csv('2021CCF用户train_data.csv',encoding = 'gb2312') 
df_test = pd.read_csv('2021CCF用户A_test_data.csv',encoding = 'gb2312')
data=pd.concat([df_train,df_test],axis=0)
for col in data.columns:
    if col not in ['ret','time','id']:
        le = LabelEncoder()
        data[col] = le.fit_transform(data[col])
data['time'] = pd.to_datetime(data['time'],format='%Y-%m-%d')
data['month']=data['time'].dt.month
data['day']=data['time'].dt.day
data['weekday']=data['time'].dt.weekday
train=data[data['ret'].notnull()]
test=data[data['ret'].isnull()]
feature=[x for x in train.columns if x not in ['ret', 'time', 'id']]
#lgb
clf = lgb.LGBMRegressor(
    learning_rate=0.05,
    n_estimators=50230,
#    num_leaves=31,
    max_depth=7,
    subsample=0.8,
#    colsample_bytree=0.8,
    metric='rmse'
)
train_x=train[feature]
target=train['ret']
test_x=test[feature]
oof1 = np.zeros(len(train))
answers = []
score = 0
n_fold = 5
folds = KFold(n_splits=n_fold, shuffle=True,random_state=2000)  
for fold_n, (train_index, valid_index) in enumerate(folds.split(train_x)):
    X_train, X_valid = train_x.iloc[train_index], train_x.iloc[valid_index]
    y_train, y_valid = target[train_index], target[valid_index]
    clf.fit(X_train,y_train,eval_set=[(X_valid, y_valid)],verbose=100,early_stopping_rounds=200)
    y_pre=clf.predict(X_valid)
    oof1[valid_index]=y_pre
    y_pred_valid = clf.predict(test_x)
    answers.append(y_pred_valid)
lgb_pre=sum(answers)/n_fold
print('score-----------',  (1/((math.sin(math.atan(np.sqrt(mean_squared_error(oof1,target)))))+1)))
sub=df_test[['id']]
sub['ret']=lgb_pre
sub.to_csv('2021CCF用户submit.csv',index=False)

baseline采用最简单的数据预处理方式，只是机械地将数据进行编码，没有考虑相关关系，也没有深度探索每个数据的业务意义，也没有进行数据可视化，这些都是后续优化的方向，优化的方向特别多，这里就不一一列举，希望baseline能给初学者带来帮助，当然，大佬就不用看了。。。

写在最后

本人才疏学浅，如果有理解不到位或者错误的地方请指正！