【sklearn】RF 交叉验证 袋外数据 参数学习曲线 网格搜索

from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_wine
from sklearn.model_selection import train_test_split
%matplotlib inline

ensemble.RandomForestClassifier

wine = load_wine()
Xtrain, Xtest, Ytrain, Ytest = train_test_split(wine.data, wine.target, test_size=0.3)

rfc = RandomForestClassifier(random_state=0)
rfc = rfc.fit(Xtrain, Ytrain)
score_r = rfc.score(Xtest, Ytest)
print(score_r)
# 0.9444444444444444

划分训练测试集，交叉验证cross_val_score、cross_validate

sklearn.model_selection.cross_val_score

from sklearn.model_selection import cross_val_score
import matplotlib.pyplot as plt

rfc = RandomForestClassifier(n_estimators=100)
rfc_s = cross_val_score(rfc, wine.data, wine.target, cv=10)
print(rfc_s)
''' array([1. , 0.94444444, 0.94444444, 0.94444444, 1. , 1. , 1. , 1. , 1. , 1. ]) '''

plt.plot(range(1, 11), rfc_s, label="RandomForest")
#plt.plot(range(1, 11), clf_s, label="DecisionTree")
plt.legend() #显示label
plt.show()

sklearn.model_selection.cross_validate

输出训练集和测试集上的交叉验证分数，可以观察训练时是否过/欠拟合

from sklearn.model_selection import cross_validate, KFold
# 实例化交叉验证方式 shuffle是否打乱数据
cv = KFold(n_splits=5, shuffle=True, random_state=1412)
result = cross_validate(RandomForestClassifier() # 评估器
                        ,wine.data,wine.target # 数据
                        ,cv=cv # 交叉验证模式
                        ,scoring='neg_mean_squared_error' # 评估指标mse(负值)
                        # 以上参数同cross_val_score()
                        # 以下cross_validate()特有参数
                        ,return_train_score=True # 返回训练集交叉验证分数
                        ,verbose=True # 打印进程
                        ,n_jobs=1 # 线程数 -1表示调用全部线程
                       )

''' [Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers. [Parallel(n_jobs=1)]: Done 5 out of 5 | elapsed: 0.6s finished '''

result
# 在训练集和测试集上的交叉验证分数差异小，则过拟合程度低
''' {'fit_time': array([0.13620615, 0.14765525, 0.13985324, 0.14356041, 0.13722539]), 'score_time': array([0.01013279, 0.01257372, 0.01011801, 0.0101881 , 0.01026726]), 'test_score': array([-0.02777778, -0.02777778, -0.05555556, -0. , -0.02857143]), 'train_score': array([-0., -0., -0., -0., -0.])} '''
# 一般情况下，交叉验证MSE分数数量级太大
# 可使用RMSE评估(MSE开平方)
abs(result['train_score'])**0.5
abs(result['test_score'])**0.5

# 画图观察
plt.figure(figsize=(8,6), dpi=80)
plt.plot(range(1, 6), abs(result['test_score'])**0.5, color='green',label='RandomForestTest')
plt.plot(range(1, 6), abs(result['train_score'])**0.5, color='red',label='RandomForestTrain')
plt.xticks([1,2,3,4,5])
plt.xlabel('CVcounts', fontsize=16)
plt.ylabel('RMSE', fontsize=16)
plt.legend()
plt.show()

无需划分训练测试集，使用boostrap

# oob_score = True
rfc = RandomForestClassifier(n_estimators=25, oob_score=True)
rfc = rfc.fit(wine.data, wine.target)
# 袋外数据作为测试集的得分
rfc.oob_score_
''' 0.9606741573033708 '''

# n_estimators学习曲线
scorel = []
for i in range(0, 200, 10):
    rfc = RandomForestClassifier(n_estimators=i+1
                                ,n_jobs=-1
                                ,random_state=0)
    score = cross_val_score(rfc, wine.data, wine.target, cv=10).mean()
    scorel.append(score)
print(max(scorel), (scorel.index(max(scorel))*10)+1)
plt.figure(figsize=[20,5])
plt.plot(range(1, 201, 10), scorel, label='n_estimators')
plt.legend()
plt.show()

0.9833333333333332 31

参数学习曲线

scorel = []
for i in range(20, 40):
    rfc = RandomForestClassifier(n_estimators=i
                                ,n_jobs=-1
                                ,random_state=0)
    score = cross_val_score(rfc, wine.data, wine.target, cv=10).mean()
    scorel.append(score)
print(max(scorel), (scorel.index(max(scorel))*10)+1)
plt.figure(figsize=[20,5])
plt.plot(range(20, 40), scorel, label='n_estimators')
plt.legend()
plt.show()

0.9833333333333334 21

网格搜索GridSearchCV

from sklearn.model_selection import GridSearchCV
import numpy as np
# 网格搜索 调整max_depth
param_grid = {
    'max_depth':np.arange(1, 20, 1), 'criterion':['gini', 'entropy']}
rfc = RandomForestClassifier(n_estimators=31
                            ,random_state=0)
GS = GridSearchCV(rfc, param_grid, cv=10)
GS.fit(wine.data, wine.target)
# 调了后泛化误差上升，就不调整了
GS.best_params_, GS.best_score_

({‘criterion’: ‘gini’, ‘max_depth’: 4}, 0.9833333333333332)