multi-label问题的不同metrics评估指标对比

【草稿】

其中阴影方框代表分子，白色空白方框+阴影方框代表分子

其中Jaccard和F1比较容易出错。

分析sklearn的jaccard_score如下：

''' jaccard 测试'''
from sklearn.metrics import jaccard_score, f1_score, hamming_loss, accuracy_score
import numpy as np
y_true = np.array([[0, 1, 0, 1, 0],[0, 0, 1, 0, 0],[1, 0, 0, 0 ,1],])
y_pred = np.array([[1, 0, 0, 0, 0],[0, 0, 1, 0 ,0],[1, 0, 0, 1, 1],])
j_macro = jaccard_score(y_true, y_pred, average='macro')
j_sample = jaccard_score(y_true, y_pred, average='samples')
j_micro = jaccard_score(y_true, y_pred, average='micro')

##### jaccard 算法 ######
xx= np.minimum(y_true, y_pred)
intersection = np.sum(np.minimum(y_true, y_pred), axis=0)
union = np.sum(np.maximum(y_true, y_pred), axis=0)
jaccard_per_class = intersection / union
macro_jaccard_score = np.mean(jaccard_per_class)
macro_jaccard_score

print('sklearn-jaccard-macro',j_macro)
print('jaccard algorithm',macro_jaccard_score)  # (1/2+0+1+0+1)/5=0.5
print('sklearn-jaccard-samples',j_sample) # (0+1+2/3)/3=5/9=0.55555
print('sklearn-jaccard-micro',j_micro)  # 3/7=0.428571428