Seaborn - 绘制多标签的混淆矩阵、召回、精准、F1
导入seaborn\matplotlib\scipy\sklearn等包:
import seaborn as sns from matplotlib import pyplot as plt from scipy.special import softmax from sklearn.metrics import accuracy_score, confusion_matrix, precision_score, recall_score, f1_score sns.set_theme(color_codes=True)
从dataframe中,获取y_true(真实标签)和y_pred(预测标签):
y_true = df["target"] y_pred = df['prediction']
计算验证数据整体的准确率acc、精准率precision、召回率recall、F1,使用加权模式average=‘weighted’:
# 准确率acc,精准precision,召回recall,F1 acc = accuracy_score(df["target"], df['prediction']) precision = precision_score(y_true, y_pred, average='weighted') recall = recall_score(y_true, y_pred, average='weighted') f1 = f1_score(y_true, y_pred, average='weighted') print(f'[Info] acc: {acc}, precision: {precision}, recall: {recall}, f1: {f1}')
计算混淆矩阵:
# 横坐标是真实类别数,纵坐标是预测类别数 cf_matrix = confusion_matrix(y_true, y_pred)
5类矩阵的绘制方案,混淆矩阵、百分比的混淆矩阵、召回矩阵、精准矩阵、F1矩阵:
- 混淆矩阵是计数,百分比的混淆矩阵是占比
- 召回矩阵是,每行的和是1,每行代表真实类别数,占比就是召回
- 精准矩阵是,每列的和是1,每列代表预测列表数,占比就是精准
- F1矩阵是按照 2PR/(P+R),注意为0的情况,需要补0,使用np.divide(a, b, out=np.zeros_like(a), where=(b != 0))