机器学习sklearn（91）：算法实例（48）分类（27）XGBoost（五）XGBoost的智慧（二）参数alpha，lambda

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

#使用网格搜索来查找最佳的参数组合
from sklearn.model_selection import GridSearchCV
param = {"reg_alpha":np.arange(0,5,0.05),"reg_lambda":np.arange(0,2,0.05)}
gscv = GridSearchCV(reg,param_grid = param,scoring = "neg_mean_squared_error",cv=cv)
#======【TIME WARNING：10~20 mins】======#
time0=time()
gscv.fit(Xtrain,Ytrain)
print(datetime.datetime.fromtimestamp(time()-time0).strftime("%M:%S:%f"))
gscv.best_params_
gscv.best_score_
preds = gscv.predict(Xtest)
from sklearn.metrics import r2_score，mean_squared_error as MSE
r2_score(Ytest,preds)
MSE(Ytest,preds) #网格搜索的结果有什么样的含义呢？为什么会出现这样的结果？你相信网格搜索得出的结果吗？试着用数学和你对XGB的理解来解释一下吧

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6 寻找最佳分枝：结构分数之差

 

 

 

 

 

 

 

 

 

 

 

 

 

7 让树停止生长：重要参数gamma

 

 

 

 

#======【TIME WARNING: 1 min】=======#
axisx = np.arange(0,5,0.05)
rs = []
var = []
ge = []
for i in axisx:
    reg = XGBR(n_estimators=180,random_state=420,gamma=i)
    result = CVS(reg,Xtrain,Ytrain,cv=cv)
    rs.append(result.mean())
    var.append(result.var())
    ge.append((1 - result.mean())**2+result.var())
print(axisx[rs.index(max(rs))],max(rs),var[rs.index(max(rs))])
print(axisx[var.index(min(var))],rs[var.index(min(var))],min(var))
print(axisx[ge.index(min(ge))],rs[ge.index(min(ge))],var[ge.index(min(ge))],min(ge))
rs = np.array(rs)
var = np.array(var)*0.1
plt.figure(figsize=(20,5))
plt.plot(axisx,rs,c="black",label="XGB")
plt.plot(axisx,rs+var,c="red",linestyle='-.')
plt.plot(axisx,rs-var,c="red",linestyle='-.')
plt.legend()
plt.show()

 

 

import xgboost as xgb
#为了便捷，使用全数据
dfull = xgb.DMatrix(X,y) #设定参数
param1 = {'silent':True,'obj':'reg:linear',"gamma":0}
num_round = 180
n_fold=5 #使用类xgb.cv
time0 = time()
cvresult1 = xgb.cv(param1, dfull, num_round,n_fold)
print(datetime.datetime.fromtimestamp(time()-time0).strftime("%M:%S:%f"))
#看看类xgb.cv生成了什么结果？
cvresult1
plt.figure(figsize=(20,5))
plt.grid()
plt.plot(range(1,181),cvresult1.iloc[:,0],c="red",label="train,gamma=0")
plt.plot(range(1,181),cvresult1.iloc[:,2],c="orange",label="test,gamma=0")
plt.legend()
plt.show()
#xgboost中回归模型的默认模型评估指标是什么？

 

 

param1 = {'silent':True,'obj':'reg:linear',"gamma":0,"eval_metric":"mae"}
cvresult1 = xgb.cv(param1, dfull, num_round,n_fold)
plt.figure(figsize=(20,5))
plt.grid()
plt.plot(range(1,181),cvresult1.iloc[:,0],c="red",label="train,gamma=0")
plt.plot(range(1,181),cvresult1.iloc[:,2],c="orange",label="test,gamma=0")
plt.legend()
plt.show()
#从这个图中，我们可以看出什么？
#怎样从图中观察模型的泛化能力？
#从这个图的角度来说，模型的调参目标是什么？

来看看如果我们调整 ，会发生怎样的变化：

param1 = {'silent':True,'obj':'reg:linear',"gamma":0}
param2 = {'silent':True,'obj':'reg:linear',"gamma":20}
num_round = 180
n_fold=5
time0 = time()
cvresult1 = xgb.cv(param1, dfull, num_round,n_fold)
print(datetime.datetime.fromtimestamp(time()-time0).strftime("%M:%S:%f"))
time0 = time()
cvresult2 = xgb.cv(param2, dfull, num_round,n_fold)
print(datetime.datetime.fromtimestamp(time()-time0).strftime("%M:%S:%f"))
plt.figure(figsize=(20,5))
plt.grid()
plt.plot(range(1,181),cvresult1.iloc[:,0],c="red",label="train,gamma=0")
plt.plot(range(1,181),cvresult1.iloc[:,2],c="orange",label="test,gamma=0")
plt.plot(range(1,181),cvresult2.iloc[:,0],c="green",label="train,gamma=20")
plt.plot(range(1,181),cvresult2.iloc[:,2],c="blue",label="test,gamma=20")
plt.legend()
plt.show()
#从这里，你看出gamma是如何控制过拟合了吗？

试一个分类的例子：

from sklearn.datasets import load_breast_cancer
data2 = load_breast_cancer()
x2 = data2.data
y2 = data2.target
dfull2 = xgb.DMatrix(x2,y2)
param1 = {'silent':True,'obj':'binary:logistic',"gamma":0,"nfold":5}
param2 = {'silent':True,'obj':'binary:logistic',"gamma":2,"nfold":5}
num_round = 100
time0 = time()
cvresult1 = xgb.cv(param1, dfull2, num_round,metrics=("error"))
print(datetime.datetime.fromtimestamp(time()-time0).strftime("%M:%S:%f"))
time0 = time()
cvresult2 = xgb.cv(param2, dfull2, num_round,metrics=("error")) 
print(datetime.datetime.fromtimestamp(time()-time0).strftime("%M:%S:%f"))
plt.figure(figsize=(20,5))
plt.grid()
plt.plot(range(1,101),cvresult1.iloc[:,0],c="red",label="train,gamma=0")
plt.plot(range(1,101),cvresult1.iloc[:,2],c="orange",label="test,gamma=0")
plt.plot(range(1,101),cvresult2.iloc[:,0],c="green",label="train,gamma=2")
plt.plot(range(1,101),cvresult2.iloc[:,2],c="blue",label="test,gamma=2")
plt.legend()
plt.show()