数据挖掘 - xgboost

1. 原理

2. 调参

3. 基本用法

# -*- coding: utf-8 -*-
import os
import pandas as pd
import xgboost as xgb

train = pd.read_csv("train.csv")
test = pd.read_csv("test.csv")
params = {
            'objective': 'binary:logistic', # 二分类, 输出0-1之间的概率
            'eta': 0.1,
            'colsample_bytree': 0.8,  # 特征随机采样
            'min_child_weight': 2,
            'max_depth': 15,
            'subsample': 0.8,         # 样本采样
            'alpha': 10,
            'gamma': 30,
            'lambda':50,
            'verbose_eval': True,
            'nthread': 8,
            'eval_metric': 'auc',
            'scale_pos_weight': 15,
            'seed': 1,
            'missing':-1
        }

# 训练, 并保存模型
if os.path.exists("clf.model"):
    clf = xgb.Booster(model_file = "clf.model")
else:    
    xgbTrain = xgb.DMatrix(train["all_features"], train["label"], params)
    clf = xgb.train(xgbTrain)
    clf.save_model("clf.model")

# 预测
prediction = clf.predict(xgb.DMatrix(test))

参考文献

https://zhuanlan.zhihu.com/p/25308120

http://www.cnblogs.com/zhangbojiangfeng/p/6428988.html