sklearn 笔记 metrics

1 分类

1.1  accuracy_score 分类准确率得分

在多标签分类中，此函数计算子集准确率：y_pred的标签集必须与 y_true 中的相应标签集完全匹配。

1.1.1 参数

• 二元分类中，等同于 jaccard_score

1.1.2 举例

from sklearn.metrics import accuracy_score

y_pred = [0, 2, 1, 3]
y_true = [0, 1, 2, 3]
accuracy_score(y_true, y_pred)
#0.5

accuracy_score(y_true, y_pred, normalize=False)
#2

1.2 balanced_accuracy_score 计算平衡准确率

平衡准确率用于二元和多类分类问题中处理数据不平衡的情况。它定义为每个类别上获得的召回率的平均值。

当 adjusted=False 时，最佳值为 1，最差值为 0

sklearn.metrics.balanced_accuracy_score(
    y_true, 
    y_pred, 
    *, 
    sample_weight=None, 
    adjusted=False)

1.2.1 参数

1.2.2 举例

from sklearn.metrics import balanced_accuracy_score
y_true = [0, 1, 0, 0, 1, 0]
y_pred = [0, 1, 0, 0, 0, 1]
balanced_accuracy_score(y_true, y_pred),accuracy_score(y_true, y_pred)
#(0.625, 0.6666666666666666)

'''
3/4+1/2=0.625
'''

1.3 top_k_accuracy_scoreTop-k 准确率分类得分

计算正确标签位于预测（按预测得分排序）的前 k 个标签中的次数

sklearn.metrics.top_k_accuracy_score(
    y_true, 
    y_score, 
    *, 
    k=2, 
    normalize=True, 
    sample_weight=None, 
    labels=None)

1.3.1 参数

1.3.2 举例

import numpy as np
from sklearn.metrics import top_k_accuracy_score
y_true = np.array([0, 1, 2, 2])
y_score = np.array([[0.5, 0.2, 0.2],  # 0 is in top 2
                    [0.3, 0.4, 0.2],  # 1 is in top 2
                    [0.2, 0.4, 0.3],  # 2 is in top 2
                    [0.7, 0.2, 0.1]]) # 2 isn't in top 2
top_k_accuracy_score(y_true, y_score, k=2)
#0.75

1.4 f1_score F1 分数

sklearn.metrics.f1_score(
    y_true, 
    y_pred, 
    *, 
    labels=None, 
    pos_label=1, 
    average='binary', 
    sample_weight=None, 
    zero_division='warn')

1.4.1  参数

1.4.2 举例

import numpy as np
from sklearn.metrics import f1_score
y_true = [0, 1, 2, 0, 1, 2]
y_pred = [0, 2, 1, 0, 0, 1]
f1_score(y_true, y_pred, average='macro')
#0.26666666666666666
f1_score(y_true, y_pred, average='micro')
#0.3333333333333333
f1_score(y_true, y_pred, average='weighted')
#0.26666666666666666
f1_score(y_true, y_pred, average=None)
#array([0.8, 0. , 0. ])

这个是怎么算出来的呢？

1.5 precision_score & recall_score

参数和用法和f1_score一致，就不多说了

1.6 jaccard_score

参数和f1_score一致

1.6 log loss 交叉熵损失

对于具有真实标签 y 和概率估计 p 的单个样本，log loss 为：

1.6.1 参数

1.6.2 举例

from sklearn.metrics import log_loss
log_loss(["spam", "ham", "ham", "spam"],
         [[.1, .9], [.9, .1], [.8, .2], [.35, .65]])
#0.21616187468057912

怎么算出来的呢？

首先，这里没有提供labels，所以需要自己从y_true中infer出来，这边有两个不同的label：“ham"和”spam"，通过排序，可以知道每一个二元组中的第一个元素是ham的概率，第二个元素是spam的概率’‘

import math

# 计算第一个样本的 log loss
loss1 = -(1 * math.log(0.9) + 0 * math.log(0.1))

# 计算第二个样本的 log loss
loss2 = -(0 * math.log(0.1) + 1 * math.log(0.9))

# 计算第三个样本的 log loss
loss3 = -(0 * math.log(0.2) + 1 * math.log(0.8))

# 计算第四个样本的 log loss
loss4 = -(1 * math.log(0.65) + 0 * math.log(0.35))

# 计算平均 log loss
average_log_loss = (loss1 + loss2 + loss3 + loss4) / 4
average_log_loss
#0.21616187468057912

2 聚类

2.1 调整互信息

sklearn.metrics.adjusted_mutual_info_score(
    labels_true, 
    labels_pred, 
    *, 
    average_method='arithmetic')

• 调整互信息（AMI，Adjusted Mutual Information）是互信息（MI，Mutual Information）的一个改进版本，用于考虑随机性的影响。
• 互信息通常会随着聚类数量的增加而增加，无论是否实际上共享了更多的信息。
• AMI通过修正互信息，排除了这种数量上的偏差。
• 对于两个聚类 U 和 V，AMI的计算公式如下：
  • 

2.1.1 参数

2.1.2 举例

from sklearn.metrics import adjusted_mutual_info_score
labels_true = [0, 0, 1, 1, 2, 2]
labels_pred = [0, 0, 1, 2, 2, 2]
ami_score = adjusted_mutual_info_score(labels_true, labels_pred)
print("Adjusted Mutual Information (AMI) score:", ami_score)
#Adjusted Mutual Information (AMI) score: 0.5023607027202738

3 回归

3.1 max_error 

sklearn.metrics.max_error(y_true, y_pred)

from sklearn.metrics import max_error
y_true = [3, 2, 17, 1]
y_pred = [4, 2, 7, 1]
max_error(y_true, y_pred)
#10

 3.2 mean_absolute_error

sklearn.metrics.mean_absolute_error(
    y_true, 
    y_pred, 
    *, 
    sample_weight=None, 
    multioutput='uniform_average')

3.2.1 参数

3.2.2  举例

from sklearn.metrics import mean_absolute_error
y_true = [0.5, -1, 7]
y_pred = [0, -1, 8]
mean_absolute_error(y_true, y_pred)
#0.5





y_true = [[0.5, 1], [-1, 1], [7, -6]]
y_pred = [[0, 2], [-1, 2], [8, -5]]
mean_absolute_error(y_true, y_pred, multioutput='raw_values')
#array([0.5, 1. ])


mean_absolute_error(y_true, y_pred)
#0.75
#上式两个取平均


mean_absolute_error(y_true, y_pred, multioutput=[0.3, 0.7])
#0.85
#0.5*0.3+1*0.7

3.3mean_squared_error

sklearn.metrics.mean_squared_error(
    y_true, 
    y_pred, 
    *, 
    sample_weight=None, 
    multioutput='uniform_average', 
    squared='deprecated')

3.4 root_mean_squared_error

3.5mean_squared_log_error

sklearn.metrics.mean_squared_log_error(
    y_true,     
    y_pred, 
    *, 
    sample_weight=None, 
    multioutput='uniform_average', 
    squared='deprecated')

3.6  root_mean_squared_log_error

root_mean_squared_log_error(
    y_true, 
    y_pred, 
    *, 
    sample_weight=None, 
    multioutput='uniform_average')

3.7  median_absolute_error

sklearn.metrics.median_absolute_error(
    y_true, 
    y_pred, 
    *, 
    multioutput='uniform_average', 
    sample_weight=None)

3.8 median_absolute_error

sklearn.metrics.median_absolute_error(
    y_true, 
    y_pred, 
    *, 
    multioutput='uniform_average', 
    sample_weight=None)

absolute error的中位数

3.9 mean_absolute_percentage_error

sklearn.metrics.mean_absolute_percentage_error(
    y_true, 
    y_pred, 
    *,     
    sample_weight=None, 
    multioutput='uniform_average')