数据挖掘竞赛lightgbm通过求最大auc调参
2024-05-10 22:50:41
一、使用步骤
0.首先展示最后的结果
# 参数含义
# learning_rate 一般设置在0.05-0.1之间
# n_estimators 100-1000 boosting的迭代次数
# min_split_gain 0 执行节点分裂的最小增益 不建议去调整
# min_child_sample 一个叶子上的最小数据量,默认设置为20。根据数据量来确定，当数据量比较大时，应提升这个数值，
# 让叶子节点的数据分布相对稳定。
# min_child_weight 一个叶子上的最小hessian和。默认设置为0.001，一般设置为1。# 需要算法调节的参数
# max_depth 树模型的最大深度。防止过拟合的最重要的参数，一般限制为3~5之间
# num_leaves 一棵树上的叶子节点个数。默认设置为31，和max_depth配合来空值树的形状，一般设置为(0, 2^max_depth - 1]的一个数值。
# subsample 默认设置为1，一般设置为0。8~1.0之间，防止过拟合。
# colsample_bytree 一般设置为0.8~1.0之间，防止过拟合
# reg_alpha L1正则化参数
# reg_lambda L2正则化参数 较大的数值会让各个特征对模型的影响力趋于均匀def roc_auc_plot(clf,x_train,y_train,x_test, y_test):train_auc = roc_auc_score(y_train,clf.predict_proba(x_train)[:,1])train_fpr, train_tpr, _ = roc_curve(y_train,clf.predict_proba(x_train)[:,1])train_ks = abs(train_fpr-train_tpr).max()print('train_ks = ', train_ks)print('train_auc = ', train_auc)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_fpr, test_tpr, _ = roc_curve(y_test,clf.predict_proba(x_test)[:,1])test_ks = abs(test_fpr-test_tpr).max()print('test_ks = ', test_ks)print('test_auc = ', test_auc)from matplotlib import pyplot as pltplt.plot(train_fpr,train_tpr,label = 'train_roc')plt.plot(test_fpr,test_tpr,label = 'test_roc')plt.plot([0,1],[0,1],'k--', c='r')plt.xlabel('False positive rate')plt.ylabel('True positive rate')plt.title('ROC Curve')plt.legend(loc = 'best')plt.show()lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=0.04,min_child_samples=68,min_child_weight=0.01,max_depth=4,num_leaves=16,colsample_bytree=0.8,subsample=0.8,reg_alpha=0.7777777777777778,reg_lambda=0.3,objective='binary')clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)
roc_auc_plot(clf,x_train,y_train,x_test, y_test)
1.读入清洗后的数据集
代码如下：
import pandas as pd
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.metrics import roc_auc_score,roc_curve,auc
import lightgbm as lgb
from multiprocessing import cpu_count
from sklearn.model_selection import KFold
import time
from sklearn.model_selection import GridSearchCV, StratifiedKFold, cross_val_score
import numpy as npdf_Master_clean = pd.read_csv(r'F:\教师培训\ppd7\df_Master_clean.csv', encoding='gb18030')
c=['Idx','target','ListingInfo','sample_status']
x = df_Master_clean[df_Master_clean['target'].notnull()].drop(columns=c)
y = df_Master_clean[df_Master_clean['target'].notnull()]['target']
x_train,x_test, y_train, y_test = train_test_split(x,y,random_state=2,test_size=0.2)
2.参数调整
代码如下（示例）：
#随机设置一组参数为基础
lgb_model1 = lgb.LGBMClassifier(n_estimators=20,boosting_type='gbdt',learning_rate=0.01,min_child_samples=100,min_child_weight=0.003,max_depth=4,num_leaves=100,colsample_bytree=0.7,subsample=0.6,reg_alpha=0.03,reg_lambda=0.3,objective='binary',max_features = 140,)
import matplotlib.pyplot as plt# 通过观察auc大小调参，找到最佳n_estimators
axisx = range(100,1001,10)
test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=i,boosting_type='gbdt',learning_rate=0.01,min_child_samples=100,min_child_weight=0.003,max_depth=4,num_leaves=100,colsample_bytree=0.7,subsample=0.6,reg_alpha=0.03,reg_lambda=0.3,objective='binary',max_features = 140,)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.xticks(axisx)
plt.legend()
plt.show()
import matplotlib.pyplot as plt# 通过观察auc大小调参，找到最佳learning_rate,learning_rate=0.04, max_auc=0.7720493433228038
axisx = np.linspace(0.01,0.05,5)test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=i,min_child_samples=100,min_child_weight=0.003,max_depth=4,num_leaves=100,colsample_bytree=0.7,subsample=0.6,reg_alpha=0.03,reg_lambda=0.3,objective='binary',max_features = 140,)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
print('learning_rate={}, max_auc={}'.format(axisx[test_aucs.index(max(test_aucs))], max(test_aucs)))plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.ylabel('auc')
plt.xticks(axisx)
plt.legend()
plt.show()
# 通过观察auc大小调参，找到最佳min_child_samples min_child_samples=68, max_auc=0.7754420865945606
# axisx = np.linspace(0.01,0.05,5)
axisx = range(65,75,1)
test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=0.04,min_child_samples=i,min_child_weight=0.003,max_depth=4,num_leaves=100,colsample_bytree=0.7,subsample=0.6,reg_alpha=0.03,reg_lambda=0.3,objective='binary',max_features = 140,)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
print('min_child_samples={}, max_auc={}'.format(axisx[test_aucs.index(max(test_aucs))], max(test_aucs)))
# test_fpr, test_tpr, _ = roc_curve(y_test,clf.predict_proba(x_test)[:,1])
# test_ks = abs(test_fpr-test_tpr).max()
# print('test_ks = ', test_ks)
# print('test_auc = ', test_auc)plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.ylabel('auc')
plt.xticks(axisx)
plt.legend()
plt.show()
# 通过观察auc大小调参，min_child_weight对auc面积没有影响
axisx = np.linspace(0.001,0.005,10)
# axisx = range(65,75,1)
# axisx = [0.001, 0.005]
test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=0.04,min_child_samples=68,min_child_weight=i,max_depth=4,num_leaves=100,colsample_bytree=0.7,subsample=0.6,reg_alpha=0.03,reg_lambda=0.3,objective='binary',max_features = 140,)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
print('min_child_weight={}, max_auc={}'.format(axisx[test_aucs.index(max(test_aucs))], max(test_aucs)))
# test_fpr, test_tpr, _ = roc_curve(y_test,clf.predict_proba(x_test)[:,1])
# test_ks = abs(test_fpr-test_tpr).max()
# print('test_ks = ', test_ks)
# print('test_auc = ', test_auc)plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.ylabel('auc')
plt.xticks(axisx)
plt.legend()
plt.show()
# 通过观察auc大小调参max_depth, max_depth=4, max_auc=0.7754420865945606
# axisx = np.linspace(0.001,0.005,10)
# axisx = range(65,75,1)
axisx = [3, 4, 5, 6, 7]
test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=0.04,min_child_samples=68,min_child_weight=0.01,max_depth=i,num_leaves=100,colsample_bytree=0.7,subsample=0.6,reg_alpha=0.03,reg_lambda=0.3,objective='binary',max_features = 140,)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
print('max_depth={}, max_auc={}'.format(axisx[test_aucs.index(max(test_aucs))], max(test_aucs)))
# test_fpr, test_tpr, _ = roc_curve(y_test,clf.predict_proba(x_test)[:,1])
# test_ks = abs(test_fpr-test_tpr).max()
# print('test_ks = ', test_ks)
# print('test_auc = ', test_auc)plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.ylabel('auc')
plt.xticks(axisx)
plt.legend()
plt.show()
# 通过观察auc大小调参num_leaves, num_leaves=16, max_auc=0.7754420865945606
# axisx = np.linspace(0.001,0.005,10)
axisx = range(10,20,1)
# axisx = [3, 4, 5, 6, 7]
test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=0.04,min_child_samples=68,min_child_weight=0.01,max_depth=4,num_leaves=i,colsample_bytree=0.7,subsample=0.6,reg_alpha=0.03,reg_lambda=0.3,objective='binary',max_features = 140,)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
print('num_leaves={}, max_auc={}'.format(axisx[test_aucs.index(max(test_aucs))], max(test_aucs)))
# test_fpr, test_tpr, _ = roc_curve(y_test,clf.predict_proba(x_test)[:,1])
# test_ks = abs(test_fpr-test_tpr).max()
# print('test_ks = ', test_ks)
# print('test_auc = ', test_auc)plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.ylabel('auc')
plt.xticks(axisx)
plt.legend()
plt.show()
# 通过观察auc大小调参colsample_bytree, colsample_bytree=0.8, max_auc=0.7770645958950596
axisx = np.linspace(0.8,0.85,5)
# axisx = range(10,101,10)
# axisx = [3, 4, 5, 6, 7]
test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=0.04,min_child_samples=68,min_child_weight=0.01,max_depth=4,num_leaves=16,colsample_bytree=i,subsample=0.6,reg_alpha=0.03,reg_lambda=0.3,objective='binary',max_features = 140,)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
print('colsample_bytree={}, max_auc={}'.format(axisx[test_aucs.index(max(test_aucs))], max(test_aucs)))
# test_fpr, test_tpr, _ = roc_curve(y_test,clf.predict_proba(x_test)[:,1])
# test_ks = abs(test_fpr-test_tpr).max()
# print('test_ks = ', test_ks)
# print('test_auc = ', test_auc)plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.ylabel('auc')
plt.xticks(axisx)
plt.legend()
plt.show()
# 通过观察auc大小调参subsample, 此参数没有对auc产生影响
axisx = np.linspace(0.7,0.8,10)
# axisx = range(10,101,10)
# axisx = [3, 4, 5, 6, 7]
test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=0.04,min_child_samples=68,min_child_weight=0.01,max_depth=4,num_leaves=16,colsample_bytree=0.8,subsample=i,reg_alpha=0.03,reg_lambda=0.3,objective='binary',)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
print('subsample={}, max_auc={}'.format(axisx[test_aucs.index(max(test_aucs))], max(test_aucs)))
# test_fpr, test_tpr, _ = roc_curve(y_test,clf.predict_proba(x_test)[:,1])
# test_ks = abs(test_fpr-test_tpr).max()
# print('test_ks = ', test_ks)
# print('test_auc = ', test_auc)plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.ylabel('auc')
plt.xticks(axisx)
plt.legend()
plt.show()
# 通过观察auc大小调参reg_alpha, reg_alpha=0.7777777777777778, max_auc=0.7773465038084242
axisx = np.linspace(0.7,0.8,10)
# axisx = range(0,1,10)
# axisx = [3, 4, 5, 6, 7]
test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=0.04,min_child_samples=68,min_child_weight=0.01,max_depth=4,num_leaves=16,colsample_bytree=0.8,subsample=0.8,reg_alpha=i,reg_lambda=0.3,objective='binary',)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
print('reg_alpha={}, max_auc={}'.format(axisx[test_aucs.index(max(test_aucs))], max(test_aucs)))
# test_fpr, test_tpr, _ = roc_curve(y_test,clf.predict_proba(x_test)[:,1])
# test_ks = abs(test_fpr-test_tpr).max()
# print('test_ks = ', test_ks)
# print('test_auc = ', test_auc)plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.ylabel('auc')
plt.xticks(axisx)
plt.legend()
plt.show()
# 通过观察auc大小调参reg_lambda, reg_lambda=0.30000000000000004, max_auc=0.7773465038084242
axisx = np.linspace(0.28,0.32,5)
# axisx = range(0,1,10)
# axisx = [3, 4, 5, 6, 7]
test_aucs = []
for i in axisx:lgb_model = lgb.LGBMClassifier(n_estimators=800,boosting_type='gbdt',learning_rate=0.04,min_child_samples=68,min_child_weight=0.01,max_depth=4,num_leaves=16,colsample_bytree=0.8,subsample=0.8,reg_alpha=0.7777777777777778,reg_lambda=i,objective='binary',)clf = lgb_model.fit(x_train, y_train,eval_set=[(x_train, y_train),(x_test,y_test)],eval_metric='auc',early_stopping_rounds=100)test_auc = roc_auc_score(y_test,clf.predict_proba(x_test)[:,1])test_aucs.append(test_auc)
print('reg_lambda={}, max_auc={}'.format(axisx[test_aucs.index(max(test_aucs))], max(test_aucs)))
# test_fpr, test_tpr, _ = roc_curve(y_test,clf.predict_proba(x_test)[:,1])
# test_ks = abs(test_fpr-test_tpr).max()
# print('test_ks = ', test_ks)
# print('test_auc = ', test_auc)plt.figure(figsize=(20,5))
plt.plot(axisx, test_aucs,c="red",label="LGBC")
plt.ylabel('auc')
plt.xticks(axisx)
plt.legend()
plt.show()
总结
我们观察整个调参过程，有的参数并没有对auc产生影响，有的效果很明显，总的来说调参后的效果要好些。
数据挖掘竞赛lightgbm通过求最大auc调参相关推荐
- 数据挖掘-二手车价格预测 Task04：建模调参
数据挖掘-二手车价格预测 Task04:建模调参 模型调参部分 利用xgb进行五折交叉验证查看模型的参数效果 ## xgb-Model xgr = xgb.XGBRegressor(n_estimat ...
- 零基础数据挖掘入门系列(五) - 模型建立与调参
思维导图:零基础入门数据挖掘的学习路径 1. 写在前面 零基础入门数据挖掘是记录自己在Datawhale举办的数据挖掘专题学习中的所学和所想, 该系列笔记使用理论结合实践的方式,整理数据挖掘相关知识, ...
- 【数据挖掘】心跳信号分类预测 之 建模调参 —— 学习笔记（四）
目录 四.建模调参 4.1 内容简介 4.2 模型原理与性质概述 4.2.1 分类模型 4.2.2 时间序列模型 4.2.3 推荐教材 4.3 模型对比与性能评估 4.3.1 逻辑回归 4.3.2 决 ...
- 数据挖掘流程（四）：建模调参
建模效果不好原因: model 模型 调参.调参可以影响5%的模型准确率. data 不同列差异过大.不同量纲可能导致模型效果不好-->取log/归一化. 同列差距过大,离散化.建模效果肯定不好 ...
- 【算法竞赛学习】心跳信号分类预测-建模与调参
Task4 建模与调参 此部分为零基础入门数据挖掘之心电图分类的 Task4 建模调参部分,带你来了解各种模型以及模型的评价和调参策略,欢迎大家后续多多交流. 赛题:零基础入门数据挖掘 - 心电图分类 ...
- Datawhale 零基础入门数据挖掘- 建模与调参
Task4 建模与调参 此部分为零基础入门数据挖掘之心电图分类的 Task4 建模调参部分,带你来了解各种模型以及模型的评价和调参策略,欢迎大家后续多多交流. 赛题:零基础入门数据挖掘 - 心电图分类 ...
- 数据挖掘竞赛-北京PM2.5浓度回归分析训练赛
北京PM2.5浓度回归分析训练赛 简介 DC上的一个回归题,比较简单. 时间原因没有细看,提交到70多名就结束了. 使用stacking方法结合多个回归模型. 过程 数据获取 官方给定. 数据探索 训 ...
- 基于集成学习方法Random Forest、Adaboost、GBDT、LightGBM、XGBoost的调参、建模、评估实现kaggle竞赛员工离职案例分析（2）
基于集成学习方法Random Forest.Adaboost.GBDT.LightGBM.XGBoost的调参.建模.评估实现kaggle竞赛员工离职案例分析(2) 引言 3. adaboost模型分 ...
- 数据挖掘小白系列!LightGBM详解与调参
1理解 [白话机器学习]算法理论+实战之LightGBM算法 备注:看了N个这个算是最详细最容易懂的了,并且有调参步骤与实例需要对XGB了解 2调参 第一步:学习率&迭代次数&估计器 ...
最新文章
- 教学思路SQL之入门习题《学员成绩》 一、建表
- RocketMQ中的主从复制
- java编写简单计算器_编写java程序简单计算器
- 【开源组件】一份值得收藏的的 MySQL 规范
- 资深Android开发带你入门Framework，架构师必备技能
- 死磕 java同步系列之开篇
- 【转】3:C#异步WaitAll的使用
- 如何将索引碎片数量降至最低
- extjs mysql_ExtJS与PHP、MySQL实现存储
- 微信小程序反编译获取前端代码
- 从铸剑到御剑：滴滴工程效能平台建设之路
- 一位清华在校生的报告
- 并查集【算法笔记/晴神笔记】
- 【信号处理】Python实现BPSK、QPSK、8PSK、8QAM、16QAM、64QAM的调制和解调
- 电脑无法复制粘贴怎么办的解决方法
- 最全Pycharm教程（40）——Pycharm扩展功能之捆绑插件TextMate
- 中兴光猫不拆机获取超级密码
- TriCore 与 RT-Thread（TC264 移植）
- 【调试】你是一名优秀的侦探吗？
- 哈密顿算子在直角坐标系的矩阵表示
热门文章
- 一起来复习Data Science：统计学中那些让人容易遗忘的知识点
- comsol计算机模拟过程,借助数值模拟分析多孔结构
- 今天我是如何10分钟完成30万售后单问题分类的
- U盘制作启动盘后只显示容量4M的问题解决
- cad自动运行dvb lisp_CAD-如何自动加载dvb、vlx、fas、lsp、arx文件
- python编写超市销售系统_Python基础项目：超市商品销售管理系统
- 51 nod 最大公约数之和 （求1~n里面各个数的gcd和 n=1000000）
- google vr 入门之制作简易的VR播放器(三)，真牛皮
- CAD Converter转PDF的使用
- Rog 幻14 2020 安装ubuntu20.04/一些小问题和美化