我試圖在 OOP 中編寫一個(gè)表，該表將返回一些算法的統(tǒng)計(jì)指標(biāo)，并在 Python 中的 pandas DataFrame 中顯示。我遇到的問題是，對于每個(gè)實(shí)例，列的名稱都會在正在創(chuàng)建的 DataFrame 中附加一個(gè)“預(yù)測”的附加字符串（最后的示例）。我的代碼：from sklearn.metrics import roc_auc_score, accuracy_score, cohen_kappa_score, recall_score, accuracy_score, precision_score, f1_scorefrom sklearn import metrics#----------------------------------------------------------------# ####################### ROC metrics table #########################----------------------------------------------------------------# class roc_table: def __init__(self, data): self.data = data def viewer(): #count columns in dataframe count_algo = len(data.columns) for i in data.iloc[:,1:]: data['predicted_{}'.format(i)] = (data[i] >= threshold).astype('int') rock_table = { "AUC":[round(roc_auc_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]], "Accuracy":[round(accuracy_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]], "Kappa":[round(cohen_kappa_score(data.actual_label, data[i]),2)for i in data.iloc[:,count_algo:]], "Sensitivity (Recall)": [round(recall_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]], "Specificity": [round(accuracy_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]], "Precision": [round(precision_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]], "F1": [round(f1_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]] } rock_table = pd.DataFrame.from_dict(rock_table, orient = 'index').reset_index() col = ['metrics'] col.extend([x for x in data.iloc[:,count_algo:]]) rock_table.columns = col return rock_table這條線給我?guī)砹寺闊篺or i in data.iloc[:,1:]: data['predicted_{}'.format(i)] = (data[i] >= threshold).astype('int')運(yùn)行它時(shí)得到的輸出示例：
1 回答
有只小跳蛙
TA貢獻(xiàn)1824條經(jīng)驗(yàn) 獲得超8個(gè)贊
問題在于您的 OOP 實(shí)現(xiàn)。您正在改變傳遞給“ roc_table ”類的原始數(shù)據(jù)。
請嘗試以下方法：
class roc_table:
def __init__(self, data):
self.org_data = data
def viewer(self, threshold):
#make a copy of initial data
data = self.org_data.copy()
#count columns in dataframe
count_algo = len(data.columns)
for i in data.iloc[:,1:]:
data['predicted_{}'.format(i)] = (data[i] >= threshold).astype('int')
rock_table = {
"AUC":[round(roc_auc_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]],
"Accuracy":[round(accuracy_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]],
"Kappa":[round(cohen_kappa_score(data.actual_label, data[i]),2)for i in data.iloc[:,count_algo:]],
"Sensitivity (Recall)": [round(recall_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]],
"Specificity": [round(accuracy_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]],
"Precision": [round(precision_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]],
"F1": [round(f1_score(data.actual_label, data[i]),2) for i in data.iloc[:,count_algo:]]
}
rock_table = pd.DataFrame.from_dict(rock_table, orient = 'index').reset_index()
col = ['metrics']
col.extend([x for x in data.iloc[:,count_algo:]])
rock_table.columns = col
return rock_table
然后像這樣實(shí)例化類并使用：
rt = roc_table(data)
threshold=0.5
rt.viewer(threshold)
threshold=0.75
rt.viewer(threshold)
這樣原始數(shù)據(jù)就不會發(fā)生變異。
希望這可以幫助。
添加回答
舉報(bào)
0/150
提交
取消