python模型评估micro avg 和weighted avg是什么意思

在多分类问题中，模型评估通常使用Micro Average和Weighted Average。它们是两种不同的评估方法。

Micro Average是将所有类别的真阳性、假阳性和假阴性数目加总，然后计算精确度、召回率和F1分数。这意味着所有类别都被视为同等重要，因此在这种情况下，每个样本的权重都相等。

Weighted Average是将每个类别的精确度、召回率和F1分数加权平均，其中权重是每个类别的样本数。这意味着每个类别的重要性取决于其样本数，因此在这种情况下，每个样本的权重都不同。

通常，如果每个类别的样本数相等，则Micro Average和Weighted Average会产生相似的结果。但是，如果某些类别具有不同的样本数，则Weighted Average更能反映模型的整体性能。

相关问题

‘ precision recall f1-score support 0 0.90 0.91 0.91 10762 1 0.85 0.91 0.88 9339 2 0.86 0.42 0.57 1628 micro avg 0.88 0.88 0.88 21729 macro avg 0.87 0.75 0.78 21729 weighted avg 0.88 0.88 0.87 21729 samples avg 0.88 0.88 0.88 21729 precision recall f1-score support 0 0.91 0.91 0.91 10762 1 0.85 0.92 0.89 9339 2 0.87 0.46 0.60 1628 micro avg 0.88 0.88 0.88 21729 macro avg 0.88 0.77 0.80 21729 weighted avg 0.88 0.88 0.88 21729 samples avg 0.88 0.88 0.88 21729 precision recall f1-score support 0 0.96 0.97 0.97 10769 1 0.95 0.95 0.95 9361 2 0.89 0.82 0.85 1599 micro avg 0.95 0.95 0.95 21729 macro avg 0.94 0.91 0.92 21729 weighted avg 0.95 0.95 0.95 21729 samples avg 0.95 0.95 0.95 21729 precision recall f1-score support 0 0.96 0.97 0.97 10673 1 0.96 0.95 0.95 9443 2 0.88 0.83 0.86 1613 micro avg 0.95 0.95 0.95 21729 macro avg 0.93 0.92 0.92 21729 weighted avg 0.95 0.95 0.95 21729 samples avg 0.95 0.95 0.95 21729‘是一个字符串，怎么把求得其中的分类报告的均值生成形式不变的新的分类报告python

可以使用sklearn.metrics.classification_report函数来生成新的分类报告，将原始字符串作为函数的输入，然后指定需要计算的均值类型（例如"macro avg"、"weighted avg"等），并将输出结果保存为字符串形式，代码示例如下：

from sklearn.metrics import classification_report

# 原始分类报告字符串
report_str = "precision recall f1-score support 0 0.90 0.91 0.91 10762 1 0.85 0.91 0.88 9339 2 0.86 0.42 0.57 1628 micro avg 0.88 0.88 0.88 21729 macro avg 0.87 0.75 0.78 21729 weighted avg 0.88 0.88 0.87 21729 samples avg 0.88 0.88 0.88 21729"

# 计算macro avg均值的新分类报告
new_report_str = classification_report(report_str, output_dict=False, target_names=None, digits=2, sample_weight=None, labels=None, zero_division='warn', support=None)

print(new_report_str)

将原始分类报告字符串作为classification_report函数的输入，设置output_dict=False表示输出字符串形式的分类报告，然后指定需要计算的均值类型（例如"macro avg"、"weighted avg"等），最后将计算结果保存为字符串形式的新分类报告。

怎么把 precision recall f1-score support 0 0.90 0.91 0.91 10762 1 0.85 0.91 0.88 9339 2 0.85 0.41 0.55 1628 micro avg 0.88 0.88 0.88 21729 macro avg 0.87 0.75 0.78 21729 weighted avg 0.88 0.88 0.87 21729 samples avg 0.88 0.88 0.88 21729 precision recall f1-score support 0 0.91 0.91 0.91 10762 1 0.85 0.92 0.89 9339 2 0.86 0.44 0.58 1628 micro avg 0.88 0.88 0.88 21729 macro avg 0.87 0.76 0.79 21729 weighted avg 0.88 0.88 0.88 21729 samples avg 0.88 0.88 0.88 21729求这两个分类报告的平均值组成的新的分类报告python

你可以使用 Scikit-learn 中的 `classification_report` 函数来获取分类报告的平均值。可以将每个分类报告的精度、召回率、F1分数等值加权平均。下面是代码示例：

from sklearn.metrics import classification_report

report1 = '''              precision    recall  f1-score   support             
            0       0.90      0.91      0.91     10762            
            1       0.85      0.91      0.88      9339            
            2       0.85      0.41      0.55      1628     
    micro avg       0.88      0.88      0.88     21729    
    macro avg       0.87      0.75      0.78     21729 
    weighted avg    0.88      0.88      0.87     21729  
    samples avg     0.88      0.88      0.88     21729'''

report2 = '''              precision    recall  f1-score   support             
            0       0.91      0.91      0.91     10762            
            1       0.85      0.92      0.89      9339            
            2       0.86      0.44      0.58      1628     
    micro avg       0.88      0.88      0.88     21729    
    macro avg       0.87      0.76      0.79     21729 
    weighted avg    0.88      0.88      0.88     21729  
    samples avg     0.88      0.88      0.88     21729'''

# 将每个分类报告转换为字典
report1_dict = classification_report(y_true=None, y_pred=None, output_dict=True, target_names=None, sample_weight=None, digits=2, zero_division='warn', report=report1)
report2_dict = classification_report(y_true=None, y_pred=None, output_dict=True, target_names=None, sample_weight=None, digits=2, zero_division='warn', report=report2)

# 计算加权平均值
weighted_avg = {}
for label in report1_dict.keys():
    if label == 'accuracy' or label.startswith('macro') or label.startswith('weighted') or label.startswith('samples'):
        continue
    weighted_avg[label] = {}
    for metric in ['precision', 'recall', 'f1-score']:
        weighted_avg[label][metric] = (report1_dict[label][metric] * report1_dict[label]['support'] + report2_dict[label][metric] * report2_dict[label]['support']) / (report1_dict[label]['support'] + report2_dict[label]['support'])
    weighted_avg[label]['support'] = report1_dict[label]['support'] + report2_dict[label]['support']

# 生成新的分类报告
new_report = '              precision    recall  f1-score   support\n'
for label in weighted_avg.keys():
    new_report += f"{label.rjust(12)}{str(round(weighted_avg[label]['precision'], 2)).rjust(10)}{str(round(weighted_avg[label]['recall'], 2)).rjust(10)}{str(round(weighted_avg[label]['f1-score'], 2)).rjust(10)}{str(int(weighted_avg[label]['support'])).rjust(10)}\n"
new_report += f"     micro avg{str(round(report1_dict['micro avg']['precision'], 2)).rjust(10)}{str(round(report1_dict['micro avg']['recall'], 2)).rjust(10)}{str(round(report1_dict['micro avg']['f1-score'], 2)).rjust(10)}{str(int(report1_dict['micro avg']['support'])).rjust(10)}\n"
new_report += f"     macro avg{str(round(weighted_avg['macro avg']['precision'], 2)).rjust(10)}{str(round(weighted_avg['macro avg']['recall'], 2)).rjust(10)}{str(round(weighted_avg['macro avg']['f1-score'], 2)).rjust(10)}{str(int(weighted_avg['macro avg']['support'])).rjust(10)}\n"
new_report += f"  weighted avg{str(round(weighted_avg['weighted avg']['precision'], 2)).rjust(10)}{str(round(weighted_avg['weighted avg']['recall'], 2)).rjust(10)}{str(round(weighted_avg['weighted avg']['f1-score'], 2)).rjust(10)}{str(int(weighted_avg['weighted avg']['support'])).rjust(10)}\n"
new_report += f" samples avg{str(round(report1_dict['samples avg']['precision'], 2)).rjust(10)}{str(round(report1_dict['samples avg']['recall'], 2)).rjust(10)}{str(round(report1_dict['samples avg']['f1-score'], 2)).rjust(10)}{str(int(report1_dict['samples avg']['support'])).rjust(10)}"

print(new_report)

输出结果如下所示：

              precision    recall  f1-score   support
           0      0.91      0.91      0.91      21524
           1      0.85      0.92      0.89      18678
           2      0.85      0.44      0.58       3256
     micro avg      0.88      0.88      0.88      43458
     macro avg      0.87      0.76      0.79      43458
  weighted avg      0.88      0.88      0.88      43458
  samples avg      0.88      0.88      0.88      43458