[人工智能]机器学习：boosing复习

非原创，代码来源葁sir

import numpy as np
import pandas as pd
from pandas import Series,DataFrame

from sklearn.ensemble import AdaBoostRegressor,GradientBoostingRegressor
# Ada的回归& GBDT的回归

from sklearn.datasets import load_boston
# 波士顿房价

from sklearn.neighbors import KNeighborsRegressor

boston = load_boston()
data = boston.data
target = boston.target
feature_names = boston.feature_names

# 建立普通的knn模型进行比较
knn = KNeighborsRegressor()

from sklearn.model_selection import train_test_split

X_train,X_test,y_train,y_test = train_test_split(data, target, test_size=0.2, random_state=1)

X_train = DataFrame(data=X_train,columns=feature_names)
X_train

404 rows × 13 columns

X_train.describe().T

X_train.min() # 没有负值 可以使用区缩放发 压缩到01

CRIM         0.00632
ZN           0.00000
INDUS        0.46000
CHAS         0.00000
NOX          0.38500
RM           3.56100
AGE          2.90000
DIS          1.12960
RAD          1.00000
TAX        187.00000
PTRATIO     12.60000
B            0.32000
LSTAT        1.73000
dtype: float64

from sklearn.preprocessing import MinMaxScaler
# 区缩放法：压缩数据到01之间

mms = MinMaxScaler()
data = mms.fit_transform(X_train)

feature_names

array(['CRIM', 'ZN', 'INDUS', 'CHAS', 'NOX', 'RM', 'AGE', 'DIS', 'RAD',
       'TAX', 'PTRATIO', 'B', 'LSTAT'], dtype='<U7')

X_train = pd.DataFrame(data=data,columns=feature_names)

X_train

404 rows × 13 columns

knn.fit(X_train,y_train)

KNeighborsRegressor()

# 评判回归问题 用什么指标：score？ mae mse
from sklearn.metrics import mean_squared_error

mean_squared_error(y_train,knn.predict(X_train))

15.028524752475246

mean_squared_error(y_test,knn.predict(X_test))

204.1549686274509

实验：adaboosting

aba = AdaBoostRegressor(base_estimator=KNeighborsRegressor(),n_estimators=100)

aba.fit(X_train,y_train)

AdaBoostRegressor(base_estimator=KNeighborsRegressor(), n_estimators=100)

mean_squared_error(y_train,aba.predict(X_train))

4.755455445544554

mean_squared_error(y_test,aba.predict(X_test))

127.99668627450978

# 每一个基学习器上在样本集上的预测结果
err_list = []
for i,y_ in enumerate(aba.staged_predict(X_test)):
    err = mean_squared_error(y_test,y_)
    err_list.append(err)
    print('C{}:ERROR:{}'.format(i,err))

C0:ERROR:199.32693725490194
C1:ERROR:199.32693725490194
C2:ERROR:220.0897333333333
C3:ERROR:199.7595607843137
C4:ERROR:207.813894117647
C5:ERROR:207.82781176470584
C6:ERROR:192.8428588235294
C7:ERROR:190.94257647058825
C8:ERROR:192.6760862745098
C9:ERROR:189.0082431372549
C10:ERROR:188.7143254901961
C11:ERROR:184.35898823529416
C12:ERROR:179.3270117647059
C13:ERROR:178.614568627451
C14:ERROR:166.1954274509804
C15:ERROR:169.03231372549018
C16:ERROR:158.3203725490196
C17:ERROR:158.4866470588235
C18:ERROR:142.9087137254902
C19:ERROR:158.1416
C20:ERROR:142.4846666666667
C21:ERROR:142.4846666666667
C22:ERROR:157.30826666666667
C23:ERROR:157.38923529411767
C24:ERROR:157.35952941176473
C25:ERROR:142.61410196078435
C26:ERROR:135.35090980392155
C27:ERROR:142.68588627450984
C28:ERROR:135.77389411764707
C29:ERROR:142.68588627450984
C30:ERROR:142.04365098039216
C31:ERROR:141.93987843137253
C32:ERROR:135.87796470588236
C33:ERROR:141.93987843137253
C34:ERROR:141.82033725490197
C35:ERROR:141.95123529411765
C36:ERROR:135.73551764705883
C37:ERROR:141.9422549019608
C38:ERROR:135.90520784313725
C39:ERROR:141.4849843137255
C40:ERROR:135.90342352941178
C41:ERROR:142.09143921568628
C42:ERROR:135.91702352941178
C43:ERROR:136.00622745098042
C44:ERROR:135.89461960784314
C45:ERROR:142.01684705882352
C46:ERROR:135.91702352941178
C47:ERROR:136.18957647058824
C48:ERROR:136.2971137254902
C49:ERROR:136.19823137254903
C50:ERROR:134.7982470588235
C51:ERROR:136.41440000000003
C52:ERROR:136.29789803921568
C53:ERROR:136.4323411764706
C54:ERROR:134.9813450980392
C55:ERROR:136.3194980392157
C56:ERROR:136.40856078431375
C57:ERROR:136.41440000000003
C58:ERROR:142.32514901960786
C59:ERROR:136.48449803921568
C60:ERROR:142.32336470588237
C61:ERROR:142.43412941176473
C62:ERROR:142.40421176470588
C63:ERROR:136.38720784313728
C64:ERROR:142.4454392156863
C65:ERROR:136.4293843137255
C66:ERROR:142.46783137254903
C67:ERROR:136.48109019607844
C68:ERROR:142.5336549019608
C69:ERROR:144.18005490196077
C70:ERROR:142.50667450980393
C71:ERROR:136.4272549019608
C72:ERROR:142.50667450980393
C73:ERROR:142.5022117647059
C74:ERROR:142.5354392156863
C75:ERROR:136.4629137254902
C76:ERROR:136.31138823529412
C77:ERROR:135.0016
C78:ERROR:136.49930196078432
C79:ERROR:135.03171764705883
C80:ERROR:135.01974901960781
C81:ERROR:128.98411372549018
C82:ERROR:135.078631372549
C83:ERROR:129.0248196078431
C84:ERROR:135.078631372549
C85:ERROR:128.98411372549018
C86:ERROR:135.078631372549
C87:ERROR:128.9571333333333
C88:ERROR:128.96369803921567
C89:ERROR:128.6254862745098
C90:ERROR:129.0248196078431
C91:ERROR:128.6512274509804
C92:ERROR:128.39080392156865
C93:ERROR:128.39080392156865
C94:ERROR:128.40443529411766
C95:ERROR:128.68032549019608
C96:ERROR:128.39080392156865
C97:ERROR:128.2014862745098
C98:ERROR:128.39080392156865
C99:ERROR:127.99668627450978

# 展示误差的变化
import seaborn as sns
import matplotlib.pyplot as plt
%matplotlib inline

sns.set()

plt.plot(err_list)

[<matplotlib.lines.Line2D at 0x1e4f02ccf40>]

# 获取每一个基学习器的错误率
plt.plot(aba.estimator_errors_)

[<matplotlib.lines.Line2D at 0x1e4ee793070>]

# 每一个基学习器的权重
plt.plot(aba.estimator_weights_)

[<matplotlib.lines.Line2D at 0x1e4ee42f940>]

GBDT观察表现

from sklearn.ensemble import RandomForestRegressor

# max_depth = None 是完全生长的决策树
RandomForestRegressor()

# 看gbdt的情况 深度限制为3 max_depth=3 需要一个弱学习器
gbdt = GradientBoostingRegressor(n_estimators=100)

gbdt.fit(X_train,y_train)

GradientBoostingRegressor()

mean_squared_error(y_train,gbdt.predict(X_train))

1.7840841714565248

mean_squared_error(y_test,gbdt.predict(X_test))

159.79621357980093

特征评估结果

# aba.feature_importances_ # knn作为基学习器的时候 没有这个对象 但是可以换基学习器为逻辑斯蒂回归等

---------------------------------------------------------------------------

AttributeError                            Traceback (most recent call last)

D:\software\anaconda\lib\site-packages\sklearn\ensemble\_weight_boosting.py in feature_importances_(self)
    253             norm = self.estimator_weights_.sum()
--> 254             return (sum(weight * clf.feature_importances_ for weight, clf
    255                     in zip(self.estimator_weights_, self.estimators_))


D:\software\anaconda\lib\site-packages\sklearn\ensemble\_weight_boosting.py in <genexpr>(.0)
    253             norm = self.estimator_weights_.sum()
--> 254             return (sum(weight * clf.feature_importances_ for weight, clf
    255                     in zip(self.estimator_weights_, self.estimators_))


AttributeError: 'KNeighborsRegressor' object has no attribute 'feature_importances_'

?

The above exception was the direct cause of the following exception:


AttributeError                            Traceback (most recent call last)

~\AppData\Local\Temp/ipykernel_12168/2668898732.py in <module>
----> 1 aba.feature_importances_


D:\software\anaconda\lib\site-packages\sklearn\ensemble\_weight_boosting.py in feature_importances_(self)
    257 
    258         except AttributeError as e:
--> 259             raise AttributeError(
    260                 "Unable to compute feature importances "
    261                 "since base_estimator does not have a "


AttributeError: Unable to compute feature importances since base_estimator does not have a feature_importances_ attribute

pd.Series(data=gbdt.feature_importances_,index=feature_names).plot(kind='bar')

<AxesSubplot:>

阅读世界，共赴山海

423全民读书节，邀你共读