使用 sklearn 获得精确度和召回率答案

【问题标题】：Getting Precision and Recall using sklearn使用 sklearn 获得精确度和召回率
【发布时间】：2018-07-04 05:38:24
【问题描述】：

使用下面的代码，我有 Accuracy 。现在我正在尝试

1) 找到每个折叠的 precision 和 recall（总共 10 折叠）

2) 为precision 获取mean

3) 为recall 获取mean

这可能类似于下面的print(scores) 和print("Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))。

有什么想法吗？

import numpy as np
from sklearn import cross_validation
from sklearn import datasets
from sklearn import svm
from sklearn.model_selection import StratifiedKFold

iris = datasets.load_iris()
skf = StratifiedKFold(n_splits=10)
clf = svm.SVC(kernel='linear', C=1)
scores = cross_validation.cross_val_score(clf, iris.data, iris.target, cv=10)
print(scores)  #[ 1. 0.93333333   1.  1. 0.86666667  1.  0.93333333   1.  1.  1.]
print("Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2)) # Accuracy: 0.97 (+/- 0.09)

【问题讨论】：

标签： python machine-learning scikit-learn svm cross-validation

【解决方案1】：

这有点不同，因为cross_val_score不能计算非二分类的precision/recall，所以需要使用recision_score、recall_score并手动进行交叉验证。参数 average='micro' 计算全局精度/召回率。

import numpy as np
from sklearn import cross_validation
from sklearn import datasets
from sklearn import svm
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import precision_score, recall_score

iris = datasets.load_iris()
skf = StratifiedKFold(n_splits=10)
clf = svm.SVC(kernel='linear', C=1)

X = iris.data
y = iris.target
precision_scores = []
recall_scores = []
for train_index, test_index in skf.split(X, y):
    X_train, X_test = X[train_index], X[test_index]
    y_train, y_test = y[train_index], y[test_index]

    y_pred = clf.fit(X_train, y_train).predict(X_test)
    precision_scores.append(precision_score(y_test, y_pred, average='micro'))
    recall_scores.append(recall_score(y_test, y_pred, average='micro'))

print(precision_scores)
print("Recall: %0.2f (+/- %0.2f)" % (np.mean(precision_scores), np.std(precision_scores) * 2))
print(recall_scores)
print("Recall: %0.2f (+/- %0.2f)" % (np.mean(recall_scores), np.std(recall_scores) * 2))

【讨论】：

我的precision、recall 和accuracy 分数都完全相同[1.0, 0.93333333333333335, 1.0, 1.0, 0.8666666666666667, 1.0, 0.93333333333333335, 1.0, 1.0, 1.0] average: 0.97 (+/- 0.09)，但情况并非如此。为什么是这样？我们该如何解决？
我得到了同样的结果。我认为这是由于数据：iris 数据集太小太简单，因此您可以尝试使用更大的数据集。
我在我正在使用的实际数据集（np.array of 2163, 8719）上尝试了这组代码，但对于precision、recall 和@，我仍然收到相同的答案987654330@
有什么想法可以解决这个问题吗？
将设置更改为average='macro' 更改了precision 和recall 分数，但我不确定它是否是适当的设置

【解决方案2】：

import pandas as pd
import numpy as np
from sklearn.metrics import confusion_matrix, recall_score, precision_score, 
                            accuracy_score, f1_score,roc_auc_score
          
def binary_classification_performance(y_test, y_pred):
    tp, fp, fn, tn = confusion_matrix(y_test, y_pred).ravel()
    accuracy = round(accuracy_score(y_pred = y_pred, y_true = y_test),2)
    precision = round(precision_score(y_pred = y_pred, y_true = y_test),2)
    recall = round(recall_score(y_pred = y_pred, y_true = y_test),2)
    f1_score = round(2*precision*recall/(precision + recall),2)
    specificity = round(tn/(tn+fp),2)
    npv = round(tn/(tn+fn),2)
    auc_roc = round(roc_auc_score(y_score = y_pred, y_true = y_test),2)


    result = pd.DataFrame({'Accuracy' : [accuracy],
                         'Precision (or PPV)' : [precision],
                         'Recall (senitivity or TPR)' : [recall],
                         'f1 score' : [f1_score],
                         'AUC_ROC' : [auc_roc],
                         'Specificty (or TNR)': [specificity],
                         'NPV' : [npv],
                         'True Positive' : [tp],
                         'True Negative' : [tn],
                         'False Positive':[fp],
                         'False Negative':[fn]})
    return result


binary_classification_performance(y_test, y_pred)

【讨论】：