• 机器学习 —— 计算评估指标

    计算评估指标

    • 假设有100个数据样本，其中有正样本70个，负样本30个
    • 现在模型查出有50个正样本，其中真正的正样本是30个
    • 求：精确率precision，召回率recall, F1值，准确率Accuracy

    TP = 30
    FP = 20
    TN = 10
    FN = 40

    # 精确率（查准率）
    precision = TP / (TP + FP) = 30 / 50 = 0.6
    # 召回率（查全率）
    recall = TP / (TP + FN) = 30 / 70 = 3/7
    # F1值
    f1 = (2 * precision * recall) / (precision + recall) = 0.5
    # 准确率
    accuracy = (TN + TP) / (TN + TP + FN + FP) = 0.4

    画ROC曲线 和 计算auc值

    1. import numpy as np
    2. import pandas as pd
    3. import matplotlib.pyplot as plt
    4.  
    5. from sklearn.datasets import load_iris
    6.  
    7. data,target = load_iris(return_X_y=True)
    8.  
    9. # 二分类
    10. target2 = target[0:100].copy()
    11. data2 = data[:100].copy()
    12.

    使用LR模型

    • from sklearn.linear_model import LogisticRegression
    • from sklearn.model_selection import train_test_split
    1. from sklearn.linear_model import LogisticRegression
    2. from sklearn.model_selection import train_test_split
    3.  
    4. x_train,x_test,y_train,y_test = train_test_split(data2,target2,test_size=0.2)
    5.  
    6. lr = LogisticRegression()
    7. lr.fit(x_train,y_train)
    8.  
    9. # 预测
    10. y_pred = lr.predict(x_test)
    11. y_pred
    12. # array([0, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1])
    13.  
    14. # ROC
    15.  
    16. # metrics：评估
    17. from sklearn.metrics import roc_curve,auc

    ROC 曲线

    1. # y_true：真是结果
    2. # y_score：预测结果
    3. fpr,tpr,_ = roc_curve(y_test,y_pred)    # 返回值：fpr,tpr,thresholds
    4. # fpr：伪阳率
    5. # tpr：真阳率
    6.  
    7.  
    8. display(fpr,tpr)
    9. '''
    10. array([0., 0., 1.])
    11. array([0., 1., 1.])
    12. '''
    13.  
    14. plt.plot(fpr,tpr)

    auc

    1. auc(fpr,tpr)
    2. # 1.0

    使用交叉验证来计算auc值，平均auc值

    • from sklearn.model_selection import KFold, StratifiedKFold
    1. from sklearn.model_selection import KFold, StratifiedKFold
    2.  
    3. skf = StratifiedKFold()
    4.  
    5. data2.shape
    6. # (100, 4)
    7.  
    8. list(skf.split(data2,target2))
    9. '''
    10. [(array([10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26,
    11.          27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43,
    12.          44, 45, 46, 47, 48, 49, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70,
    13.          71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87,
    14.          88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99]),
    15.   array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 50, 51, 52, 53, 54, 55, 56,
    16.          57, 58, 59])),
    17.  (array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 20, 21, 22, 23, 24, 25, 26,
    18.          27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43,
    19.          44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 70,
    20.          71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87,
    21.          88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99]),
    22.   array([10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 60, 61, 62, 63, 64, 65, 66,
    23.          67, 68, 69])),
    24.  (array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
    25.          17, 18, 19, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43,
    26.          44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60,
    27.          61, 62, 63, 64, 65, 66, 67, 68, 69, 80, 81, 82, 83, 84, 85, 86, 87,
    28.          88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99]),
    29.   array([20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 70, 71, 72, 73, 74, 75, 76,
    30.          77, 78, 79])),
    31.  (array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
    32.          17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 40, 41, 42, 43,
    33.          44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60,
    34.          61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77,
    35.          78, 79, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99]),
    36.   array([30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 80, 81, 82, 83, 84, 85, 86,
    37.          87, 88, 89])),
    38.  (array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
    39.          17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
    40.          34, 35, 36, 37, 38, 39, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60,
    41.          61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77,
    42.          78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89]),
    43.   array([40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 90, 91, 92, 93, 94, 95, 96,
    44.          97, 98, 99]))]
    45. '''
    46.  
    47. for train,test in skf.split(data2,target2):
    48.     x_train = data2[train]
    49.     y_train = target2[train]
    50.     x_test = data2[test]
    51.     y_test = target2[test]
    52.     
    53.     
    54.     # LR
    55.     lr = LogisticRegression()
    56.     lr.fit(x_train,y_train)
    57.     y_pred = lr.predict(x_test)
    58.     
    59.     # roc
    60.     fpr,tpr,_ = roc_curve(y_test,y_pred)
    61.     plt.plot(fpr,tpr)
    62.     print(auc(fpr,tpr))
    63. '''
    64. 1.0
    65. 1.0
    66. 1.0
    67. 1.0
    68. 1.0
    69. '''

    添加噪声

    • 给data2添加500列随机值
    1. data2.shape
    2. # (100, 4)
    3.  
    4. data3 = np.random.randn(100,500)
    5. data3.shape
    6. # (100, 500)
    7.  
    8. # 左右拼接：水平拼接
    9. data4 = np.hstack((data2,data3))
    10. data4.shape
    11. # (100, 504)
    12.  
    13. skf = StratifiedKFold()
    14.  
    15. auc_list = []
    16. for train,test in skf.split(data4,target2):
    17.     x_train = data4[train]
    18.     y_train = target2[train]
    19.     x_test = data4[test]
    20.     y_test = target2[test]
    21.     
    22.     
    23.     # LR
    24.     lr = LogisticRegression()
    25.     lr.fit(x_train,y_train)
    26.     # 预测
    27.     # y_pred = lr.predict(x_test)
    28.     # 预测概率
    29.     y_proba = lr.predict_proba(x_test)
    30.     print('y_proba:',y_proba)
    31.     
    32.     
    33.     # roc
    34.     fpr,tpr,_ = roc_curve(y_test,y_proba[:,1])
    35.     
    36.     # 画图
    37.     plt.plot(fpr,tpr)
    38.  
    39.     print('fpr:',fpr)
    40.     print('tpr:',tpr)
    41.     print('auc:',auc(fpr,tpr))
    42.     print('*'*100)
    43.     
    44.     auc_list.append(auc(fpr,tpr))
    45.  
    46. # 平均 auc
    47. np.array(auc_list).mean()
    48. '''
    49. y_proba: [[0.3267921  0.6732079 ]
    50.  [0.96683557 0.03316443]
    51.  [0.77520064 0.22479936]
    52.  [0.65359444 0.34640556]
    53.  [0.28117064 0.71882936]
    54.  [0.51257663 0.48742337]
    55.  [0.89757814 0.10242186]
    56.  [0.70565166 0.29434834]
    57.  [0.95428978 0.04571022]
    58.  [0.79620831 0.20379169]
    59.  [0.11122497 0.88877503]
    60.  [0.14503562 0.85496438]
    61.  [0.09769969 0.90230031]
    62.  [0.1427527  0.8572473 ]
    63.  [0.64864805 0.35135195]
    64.  [0.77964905 0.22035095]
    65.  [0.50532259 0.49467741]
    66.  [0.88917687 0.11082313]
    67.  [0.20508718 0.79491282]
    68.  [0.22918407 0.77081593]]
    69. fpr: [0.  0.  0.  0.2 0.2 0.3 0.3 0.6 0.6 0.7 0.7 1. ]
    70. tpr: [0.  0.1 0.6 0.6 0.7 0.7 0.8 0.8 0.9 0.9 1.  1. ]
    71. auc: 0.82
    72. ****************************************************************************************************
    73. y_proba: [[0.81694936 0.18305064]
    74.  [0.58068561 0.41931439]
    75.  [0.95133392 0.04866608]
    76.  [0.40420908 0.59579092]
    77.  [0.3271581  0.6728419 ]
    78.  [0.99027305 0.00972695]
    79.  [0.64918216 0.35081784]
    80.  [0.90200046 0.09799954]
    81.  [0.63054898 0.36945102]
    82.  [0.93316453 0.06683547]
    83.  [0.53006938 0.46993062]
    84.  [0.17861305 0.82138695]
    85.  [0.006705   0.993295  ]
    86.  [0.09477154 0.90522846]
    87.  [0.56917531 0.43082469]
    88.  [0.03227622 0.96772378]
    89.  [0.22280499 0.77719501]
    90.  [0.15966529 0.84033471]
    91.  [0.02610573 0.97389427]
    92.  [0.01608401 0.98391599]]
    93. fpr: [0.  0.  0.  0.2 0.2 1. ]
    94. tpr: [0.  0.1 0.8 0.8 1.  1. ]
    95. auc: 0.9600000000000001
    96. ****************************************************************************************************
    97. y_proba: [[0.73755142 0.26244858]
    98.  [0.81486985 0.18513015]
    99.  [0.98155993 0.01844007]
    100.  [0.62469409 0.37530591]
    101.  [0.86580681 0.13419319]
    102.  [0.93865476 0.06134524]
    103.  [0.76684129 0.23315871]
    104.  [0.26828926 0.73171074]
    105.  [0.95379293 0.04620707]
    106.  [0.82872899 0.17127101]
    107.  [0.0450968  0.9549032 ]
    108.  [0.4752642  0.5247358 ]
    109.  [0.38068224 0.61931776]
    110.  [0.56844634 0.43155366]
    111.  [0.49825931 0.50174069]
    112.  [0.05526257 0.94473743]
    113.  [0.04108483 0.95891517]
    114.  [0.00417408 0.99582592]
    115.  [0.09069155 0.90930845]
    116.  [0.42708884 0.57291116]]
    117. fpr: [0.  0.  0.  0.1 0.1 1. ]
    118. tpr: [0.  0.1 0.5 0.5 1.  1. ]
    119. auc: 0.9500000000000001
    120. ****************************************************************************************************
    121. y_proba: [[0.89441894 0.10558106]
    122.  [0.65744045 0.34255955]
    123.  [0.67092317 0.32907683]
    124.  [0.78029511 0.21970489]
    125.  [0.69217484 0.30782516]
    126.  [0.97861482 0.02138518]
    127.  [0.711046   0.288954  ]
    128.  [0.94908913 0.05091087]
    129.  [0.62170149 0.37829851]
    130.  [0.57082372 0.42917628]
    131.  [0.59759391 0.40240609]
    132.  [0.53269573 0.46730427]
    133.  [0.08361238 0.91638762]
    134.  [0.3546565  0.6453435 ]
    135.  [0.13494363 0.86505637]
    136.  [0.01205661 0.98794339]
    137.  [0.04489417 0.95510583]
    138.  [0.57049956 0.42950044]
    139.  [0.3636283  0.6363717 ]
    140.  [0.13165516 0.86834484]]
    141. fpr: [0.  0.  0.  0.1 0.1 1. ]
    142. tpr: [0.  0.1 0.9 0.9 1.  1. ]
    143. auc: 0.99
    144. ****************************************************************************************************
    145. y_proba: [[0.85161531 0.14838469]
    146.  [0.9726683  0.0273317 ]
    147.  [0.53251231 0.46748769]
    148.  [0.72269431 0.27730569]
    149.  [0.87414963 0.12585037]
    150.  [0.79130481 0.20869519]
    151.  [0.98550565 0.01449435]
    152.  [0.56034861 0.43965139]
    153.  [0.55647585 0.44352415]
    154.  [0.72393126 0.27606874]
    155.  [0.03734951 0.96265049]
    156.  [0.16550755 0.83449245]
    157.  [0.28703024 0.71296976]
    158.  [0.1594562  0.8405438 ]
    159.  [0.07379419 0.92620581]
    160.  [0.48656743 0.51343257]
    161.  [0.3818963  0.6181037 ]
    162.  [0.23117614 0.76882386]
    163.  [0.4644294  0.5355706 ]
    164.  [0.46337177 0.53662823]]
    165. fpr: [0. 0. 0. 1.]
    166. tpr: [0.  0.1 1.  1. ]
    167. auc: 1.0
    168. ****************************************************************************************************
    169. 0.944
    170. '''

    线性插值

    1. x = np.linspace(0,10,30)
    2. y = np.sin(x)
    3.  
    4. plt.scatter(x,y)

    1. x2 = np.linspace(0,10,100)
    2.  
    3. # interp：线性插值
    4. # 让 x2，y2 之间的关系和 x，y之间的关系一样
    5. y2 = np.interp(x2,x,y)
    6.  
    7. plt.scatter(x,y)
    8. plt.scatter(x2,y2,marker='*')

    计算平均AUC值，和平均ROC曲线

    • auc <= 0.5 : 模型很差
    • auc > 0.6 : 模型一般
    • auc > 0.7 : 模型还可以
    • auc > 0.8 : 模型较好
    • auc > 0.9 : 模型非常好

     

    1. # 算平均AUC值
    2. np.array(auc_list).mean()
    3. # 0.944
    4.  
    5. # 相当于 x 轴
    6. fprs = np.linspace(0,1,101)
    7.  
    8. tprs_list = []
    9. auc_list = []
    10.  
    11. for train,test in skf.split(data4,target2):
    12.     x_train = data4[train]
    13.     y_train = target2[train]
    14.     x_test = data4[test]
    15.     y_test = target2[test]
    16.     
    17.     
    18.     # LR
    19.     lr = LogisticRegression()
    20.     lr.fit(x_train,y_train)
    21.     # 预测
    22.     # y_pred = lr.predict(x_test)
    23.     # 预测概率
    24.     y_proba = lr.predict_proba(x_test)
    25.     
    26.     
    27.     # roc
    28.     fpr,tpr,_ = roc_curve(y_test,y_proba[:,1])
    29.     
    30.     auc_ = auc(fpr,tpr)
    31.     auc_list.append(auc_)
    32.     
    33.     # 画图
    34.     plt.plot(fpr,tpr,ls='--',label=f'auc:{np.round(auc_,2)}')
    35.     
    36.     
    37.     # 线性插值
    38.     # 让 fprs 与 tprs 的关系和 fpr 与 tpr 的关系一样
    39.     tprs = np.interp(fprs,fpr,tpr)
    40.     
    41.     tprs_list.append(tprs)
    42.  
    43.  
    44. # 平均 tprs
    45. tprs_mean = np.array(tprs_list).mean(axis=0)
    46.  
    47. auc_mean = np.array(auc_list).mean()
    48.  
    49. # 画平均ROC图
    50. plt.plot(fprs,tprs_mean,label=f'auc_mean:{np.round(auc_mean,2)}')

  • 相关阅读:
    Android /android_vendor.32_arm64_armv8-a_shared/libtinyals a.so.abidiff报错
    【前端面试常问】MVC与MVVM
    （209）Verilog HDL：设计一个电路之Rule 90
    STM32F103C8/BT6 USART1 DMA发送失败
    【PAT】数据结构树和图月考复习1
    鸿蒙项目实战-月木学途：1.编写首页,包括搜索栏、轮播图、宫格
    elasticsearch 安装教程
    shardingsphere分库分表示例（逻辑表，真实表，绑定表，广播表，单表）
    面试十三、malloc 、calloc、realloc以及new的区别
    关于类的继承
  • 原文地址：https://blog.csdn.net/qq_52421831/article/details/127849837