xgboost学习
1、原理
https://www.cnblogs.com/zhouxiaohui888/p/6008368.html
2、实战
xgboost中比较重要的参数介绍:
(1)学习率:learning rate :一般设置比较低,0.1以下
(2)tree:
max_depth
min_child_weight
subsample
colsample_bytree
gamma
(3)正则化参数
lambda
alpha
(1)objective [ default=reg:linear ] 定义学习任务及相应的学习目标,可选的目标函数如下:
- “reg:linear” –线性回归。
- “reg:logistic” –逻辑回归。
- “binary:logistic” –二分类的逻辑回归问题,输出为概率。
- “binary:logitraw” –二分类的逻辑回归问题,输出的结果为wTx。
- “count:poisson” –计数问题的poisson回归,输出结果为poisson分布。 在poisson回归中,max_delta_step的缺省值为0.7。(used to safeguard optimization)
- “multi:softmax” –让XGBoost采用softmax目标函数处理多分类问题,同时需要设置参数num_class(类别个数)
- “multi:softprob” –和softmax一样,但是输出的是ndata * nclass的向量,可以将该向量reshape成ndata行nclass列的矩阵。没行数据表示样本所属于每个类别的概率。
- “rank:pairwise” –set XGBoost to do ranking task by minimizing the pairwise loss
(2)’eval_metric’ The choices are listed below,评估指标:
- “rmse”: root mean square error
- “logloss”: negative log-likelihood
- “error”: Binary classification error rate. It is calculated as #(wrong cases)/#(all cases). For the predictions, the evaluation will regard the instances with prediction value larger than 0.5 as positive instances, and the others as negative instances.
- “merror”: Multiclass classification error rate. It is calculated as #(wrong cases)/#(all cases).
- “mlogloss”: Multiclass logloss
- “auc”: Area under the curve for ranking evaluation.
- “ndcg”:Normalized Discounted Cumulative Gain
- “map”:Mean average precision
- “ndcg@n”,”map@n”: n can be assigned as an integer to cut off the top positions in the lists for evaluation.
- “ndcg-“,”map-“,”ndcg@n-“,”map@n-“: In XGBoost, NDCG and MAP will evaluate the score of a list without any positive samples as 1. By adding “-” in the evaluation metric XGBoost will evaluate these score as 0 to be consistent under some conditions.
(3)lambda [default=0] L2 正则的惩罚系数
(4)alpha [default=0] L1 正则的惩罚系数
(5)lambda_bias 在偏置上的L2正则。缺省值为0(在L1上没有偏置项的正则,因为L1时偏置不重要)
(6)eta [default=0.3]
为了防止过拟合,更新过程中用到的收缩步长。在每次提升计算之后,算法会直接获得新特征的权重。 eta通过缩减特征的权重使提升计算过程更加保守。缺省值为0.3
取值范围为:[0,1]
(7)max_depth [default=6] 数的最大深度。缺省值为6 ,取值范围为:[1,∞]
(8)min_child_weight [default=1]
孩子节点中最小的样本权重和。如果一个叶子节点的样本权重和小于min_child_weight则拆分过程结束。在现行回归模型中,这个参数是指建立每个模型所需要的最小样本数。该成熟越大算法越conservative
取值范围为: [0,∞]
xgb1=XGBClassifier(
learning_rate=0.1,
n_estimators=1000,
max_depth=5,
min_child_weight=1,
gamma=0,
subsample=0.8
colsample_bytree=0.8,
objective='binary:logistic',
nthread=4,
scale_pos_weight=1,
seed=27)
3、xgboost重要模块:plot_importance【显示特征的重要性】
from xgboost import XGBClassifier
from xgboost import plot_importance
from matplotlib import pyplot model=XGBClassifier()
model.fit(X,Y)
plot_importance(model)
pyplot.show()
#图中就可以显示出各种特征的重要性
xgboost学习的更多相关文章
- 【新人赛】阿里云恶意程序检测 -- 实践记录11.10 - XGBoost学习 / 代码阅读、调参经验总结
XGBoost学习: 集成学习将多个弱学习器结合起来,优势互补,可以达到强学习器的效果.要想得到最好的集成效果,这些弱学习器应当"好而不同". 根据个体学习器的生成方法,集成学习方 ...
- XGboost学习总结
XGboost,全称Extrem Gradient boost,极度梯度提升,是陈天奇大牛在GBDT等传统Boosting算法的基础上重新优化形成的,是Kaggle竞赛的必杀神器. XGboost属于 ...
- xgboost学习与总结
最近在研究xgboost,把一些xgboost的知识总结一下.这里只是把相关资源作总结,原创的东西不多. 原理 xgboost的原理首先看xgboost的作者陈天奇的ppt 英文不太好的同学可以看看这 ...
- 数据竞赛利器 —— xgboost 学习清单
1. 入门大全 xgboost 作者给出的一份完备的使用 xgboost 进行数据分析的完整示例代码:A walk through python example for UCI Mushroom da ...
- XGBoost学习笔记2
XGBoost API 参数 分类问题 使用逻辑回归 # Import xgboost import xgboost as xgb # Create arrays for the features a ...
- XGBoost学习笔记1
XGBoost XGBoost这个网红大杀器,似乎很好用,完事儿还是自己推导一遍吧,datacamp上面有辅助的课程,但是不太涉及原理 它究竟有多好用呢?我还没用过,先搞清楚原理,hahaha~ 参考 ...
- 【Python机器学习实战】决策树与集成学习(七)——集成学习(5)XGBoost实例及调参
上一节对XGBoost算法的原理和过程进行了描述,XGBoost在算法优化方面主要在原损失函数中加入了正则项,同时将损失函数的二阶泰勒展开近似展开代替残差(事实上在GBDT中叶子结点的最优值求解也是使 ...
- xgboost原理及应用
1.背景 关于xgboost的原理网络上的资源很少,大多数还停留在应用层面,本文通过学习陈天奇博士的PPT 地址和xgboost导读和实战 地址,希望对xgboost原理进行深入理解. 2.xgboo ...
- xgboost原理及应用--转
1.背景 关于xgboost的原理网络上的资源很少,大多数还停留在应用层面,本文通过学习陈天奇博士的PPT地址和xgboost导读和实战 地址,希望对xgboost原理进行深入理解. 2.xgboos ...
随机推荐
- input type=”file“ change事件只执行一次的问题
js解决办法 HTML:<input id="file",type="file" onchange="upload()" /> ...
- 原生node实现简易留言板
原生node实现简易留言板 学习node,实现一个简单的留言板小demo 1. 使用模块 http模块 创建服务 fs模块 操作读取文件 url模块 便于path操作并读取表单提交数据 art-tem ...
- ssh远程连接和linux基本操作
客户端工具:Xshell,SecureCRT 启动网卡(eth0): ifup eth0 查看IP地址: ifconfig [root@oldboy~] : [登入名 @ 主机名 文件或者 ...
- 四大MQ比较及MQ详解
消息队列已经逐渐成为企业IT系统内部通信的核心手段.它具有低耦合.可靠投递.广播.流量控制.最终一致性等一系列功能,成为异步RPC的主要手段之 一.当今市面上有很多主流的消息中间件,如老牌的Activ ...
- 2015 Multi-University Training Contest 10 hdu 5411 CRB and Puzzle
CRB and Puzzle Time Limit: 2000/1000 MS (Java/Others) Memory Limit: 65536/65536 K (Java/Others)To ...
- 机房-动环-江森ODS系统
优势: 标准的BACnet系统平台 开放的集成特性 支持Desktop and Server平台 支持多达100个NxE 支持无线应用,可以手机访问 DCIM---数据中心基础架构管理平台介绍 不同于 ...
- Mysql 下DELETE操作表别名问题
在用DELETE删除mysql数据库数据时采取一下两种方式: 方式一:DELETE FROM B_PROSON WHERE ID = 1; 不使用别名 方式二:DELETE BP FROM B_PRO ...
- HDU 4394 BFS
M2%10x=N (x=0,1,2,3....) 给出N.找到最小的满足条件的M 因为:N的个位仅仅由M的个位决定.N十位由M的个位和十位决定,N的百位由M的个位十位百位决定.以此类推 全部从个位開始 ...
- 【解决方法】UITableView 性能优化笔记
1.网络图片异步载入,SDWebImage. 2.文字直接 drawInRect/drawAtPoint 绘制,參考 ABTableViewCell.AdvancedTableViewCells. 3 ...
- [BZOJ 3884][欧拉定理]上帝与集合的正确使用方法
看看我们机房某畸形写的题解:http://blog.csdn.net/sinat_27410769/article/details/46754209 此题为popoQQQ神犇所出,在此orz #inc ...