使用sklear.metrics计算precision等指标

在Precision、Recall、F1-score、Micro-F1、Macro-F1、Recall@K文章中介绍了一些分类指标的理论，计算方式。在本文中将介绍使用sklearn.metrics库来计算这些指标

1. accuracy

sklearn.metrics.accuracy_score(y_true, y_pred, normalize=True, sample_weight=None)

y_true: 真实标签
y_pred: 预测标签
normalize: 若为True结果返回准确率，若为False结果返回分类正确的样本个数
注意：不支持多标签

    import numpy as np
    from sklearn.metrics import accuracy_score
    y_pred = [0, 2, 1, 3]
    y_true = [0, 1, 2, 3]
    rate = accuracy_score(y_true, y_pred)
    num = accuracy_score(y_true, y_pred, normalize=False)
    print("rate:", rate)
    print("num:", num)

输出

rate: 0.5
num: 2

2. 构建文本报告展示主要分类指标

sklearn.metrics.classification_report(y_true, y_pred, labels=None, target_names=None, sample_weight=None, digits=2)

labels: 标签的索引
target_names: 标签的名字（与labels对应）
digits: 小数点位数

    from sklearn.metrics import classification_report
    y_true = [0, 1, 2, 2, 2]
    y_pred = [0, 0, 2, 2, 1]
    target_names = ['class 0', 'class 1', 'class 2']
    print(classification_report(y_true, y_pred, target_names=target_names))

输出：

              precision    recall  f1-score   support
     class 0       0.50      1.00      0.67         1
     class 1       0.00      0.00      0.00         1
     class 2       1.00      0.67      0.80         3

    accuracy                           0.60         5
   macro avg       0.50      0.56      0.49         5
weighted avg       0.70      0.60      0.61         5

注意：当labels为None时，target_name中的名字默认将labels视为label_index从小到大排序，并与之对应。当指定labels参数时，target_names中名字与labels中索引对应。将上面代码最后一行改为

print(classification_report(y_true, y_pred, labels=[0, 2, 1], target_names=target_names))

输出：

              precision    recall  f1-score   support
     class 0       0.50      1.00      0.67         1
     class 1       1.00      0.67      0.80         3
     class 2       0.00      0.00      0.00         1
     
    accuracy                           0.60         5
   macro avg       0.50      0.56      0.49         5
weighted avg       0.70      0.60      0.61         5

可以看到class1和class2的输出与上面的例子换了下

3. F1-score

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

labels: 标签的索引
pos_label: 作用不大
average: 可选[None, ‘binary’ (default), ‘micro’, ‘macro’, ‘samples’, ‘weighted’]之一，如果指定binary结果只返回二分类中的正样本的分数；指定其他分别计算micro-f1，macro-f1，如果是None返回每一个类别的f1-score，是一个列表；若是weighted，先计算每个类别的f1，然后根据每个类别样本数目加权平均

from sklearn.metrics import f1_score, confusion_matrix
y_true = [0, 1, 2, 0, 1, 2]
y_pred = [0, 2, 1, 0, 0, 1]
matrix = confusion_matrix(y_true, y_pred, labels=[0, 1, 2])
macro_f1 = f1_score(y_true, y_pred, average='macro')  
micro_f1 = f1_score(y_true, y_pred, average='micro')  
weight_f1 = f1_score(y_true, y_pred, average='weighted')  
all_f1 = f1_score(y_true, y_pred, average=None)
print("matrix:", matrix)
print("macro:", macro_f1)
print("micro:", micro_f1)
print("weight:", weight_f1)
print("all:", all_f1)

matrix: [[2 0 0]
		 [1 0 1]
		 [0 2 0]]
macro: 0.26666666666666666
micro: 0.3333333333333333
weight: 0.26666666666666666
all: [0.8 0.  0. ]

4. $F_{\beta }$-score

sklearn.metrics.fbeta_score(y_true, y_pred, beta, labels=None, pos_label=1, average='binary', sample_weight=None)

$\beta$参数决定了precision的权重，$\beta <1$precison有更多的权重，反之recall有更多的权重
beta: 公式中$\beta$值
其他参数与f1-score相同

5. precision

sklearn.metrics.precision_score(y_true, y_pred, labels=None, pos_label=1, average='binary', sample_weight=None)

参数说明与f1-score相同

from sklearn.metrics import precision_score
y_true = [0, 1, 2, 0, 1, 2]
y_pred = [0, 2, 1, 0, 0, 1]
macro = precision_score(y_true, y_pred, average='macro')  
micro = precision_score(y_true, y_pred, average='micro')  
weight = precision_score(y_true, y_pred, average='weighted')
all = precision_score(y_true, y_pred, average=None)
print("macro:", macro)
print("micro:", micro)
print("weight:", weight)
print("all:", all)

macro: 0.2222222222222222
micro: 0.3333333333333333
weight: 0.2222222222222222
all: [0.66666667 0.         0.        ]

6. recall

sklearn.metrics.recall_score(y_true, y_pred, labels=None, pos_label=1, average='binary', sample_weight=None)

用法与precision相同

参考
sklearn api官方文档