如何使用 scikit learn 遍历 python 中的模型“列表”？

Question

我构建了一个函数来显示单个模型的一些评估指标，现在我想将此函数应用于我估计的模型池。

旧函数的输入是：

OldFunction(code: str, x, X_train: np.array, X_test: np.array, X:pd.DataFrame)

地点：

code 是一个字符串，用于创建数据框的列名
x 是型号名称
X_train 和 X_test 是数据拆分器的 np.arrays
X是整个数据的数据框

为了估计模型池的指标，我尝试通过在函数中添加循环并将模型放入列表来修改函数。

但它不起作用。

出现问题是因为我无法遍历模型列表，那么我有什么选择？你有什么想法吗？

我把新功能留在下面。

import numpy as np
import pandas as pd
def displaymetrics(code: list, models: list, X_train: np.array, X_test: np.array, X: pd.DataFrame):
    for i in models:
        import pandas as pd
        import numpy as np
        from sklearn.metrics import roc_curve, auc
        from sklearn.metrics import accuracy_score, recall_score, precision_score
        from sklearn.model_selection import cross_val_score

        y_score = models[i].fit(X_train, y_train).decision_function(X_test)
        fpr, tpr, _ = roc_curve(y_test, y_score)
        roc_auc = auc(fpr, tpr)

        # Traditional Scores

        y_pred = pd.DataFrame(model[i].predict(X_train)).reset_index(drop=True)
        Recall_Train,Precision_Train, Accuracy_Train  = recall_score(y_train, y_pred), precision_score(y_train, y_pred), accuracy_score(y_train, y_pred)
        y_pred = pd.DataFrame(model[i].predict(X_test)).reset_index(drop=True)
        Recall_Test = recall_score(y_test, y_pred)
        Precision_Test = precision_score(y_test, y_pred)
        Accuracy_Test = accuracy_score(y_test, y_pred)

        #Cross Validation
        cv_au = cross_val_score(models[i], X_test, y_test, cv=30, scoring='roc_auc')
        cv_f1 = cross_val_score(models[i], X_test, y_test, cv=30, scoring='f1')
        cv_pr = cross_val_score(models[i], X_test, y_test, cv=30, scoring='precision')
        cv_re = cross_val_score(models[i], X_test, y_test, cv=30, scoring='recall')
        cv_ac = cross_val_score(models[i], X_test, y_test, cv=30, scoring='accuracy')
        cv_ba = cross_val_score(models[i], X_test, y_test, cv=30, scoring='balanced_accuracy')
        cv_au_m, cv_au_std =  cv_au.mean() , cv_au.std() 
        cv_f1_m, cv_f1_std = cv_f1.mean() , cv_f1.std()
        cv_pr_m, cv_pr_std = cv_pr.mean() , cv_pr.std()
        cv_re_m, cv_re_std= cv_re.mean() , cv_re.std()
        cv_ac_m, cv_ac_std = cv_ac.mean() , cv_ac.std()
        cv_ba_m, cv_ba_std= cv_ba.mean() , cv_ba.std()
        cv_au, cv_f1, cv_pr =  (cv_au_m, cv_au_std),  (cv_f1_m, cv_f1_std), (cv_pr_m, cv_pr_std) 
        cv_re, cv_ac, cv_ba = (cv_re_m, cv_re_std), (cv_ac_m, cv_ac_std), (cv_ba_m, cv_ba_std)
        tuples = [cv_au, cv_f1, cv_pr, cv_re, cv_ac, cv_ba]
        tuplas = [0]*len(tuples)
        for i in range(len(tuples)):
            tuplas[i] = [round(x,4) for x in tuples[i]]
        results = pd.DataFrame()
        results['Metrics'] = ['roc_auc', 'Accuracy_Train', 'Precision_Train', 'Recall_Train', 'Accuracy_Test', 
                              'Precision_Test','Recall_Test', 'cv_roc-auc (mean, std)', 'cv_f1score(mean, std)', 
                              'cv_precision (mean, std)', 'cv_recall (mean, std)', 'cv_accuracy (mean, std)', 
                              'cv_bal_accuracy (mean, std)']
        results.set_index(['Metrics'], inplace=True)
        results['Model_'+code[i]] = [roc_auc, Accuracy_Train, Precision_Train, Recall_Train, Accuracy_Test, 
                            Precision_Test, Recall_Test, tuplas[0], tuplas[1], tuplas[2], tuplas[3],
                           tuplas[4], tuplas[5]]

    return results

输出应该是一个数据框，其中每列代表每个模型，行代表指标。

Answer 1

您可能应该提及是否有错误或只是输出不正确。 我会假设你有一个错误。

您确定在调用displaymetrics 时将模型作为列表传递吗？

例如

models = [model1, model2, ...]
displaymetrics(code, models, X_train, X_test, X)

另外，您的代码中存在错误： 你打电话给models[i].fit(...) 但i 本身就是一个模型。您应该只做i.fit(...) 或更好地更改名称i，因为它通常指的是迭代内容。 （如果你想遍历列表的索引，你应该使用for i in range(0, len(models)): ...。）

注意：您不应该在每次模型迭代时都导入 pandas 和 numpy。我还建议您将所有导入（sklearn 模块的）放在代码的上部。

所以，我认为你的代码应该是这样的：

import numpy as np
import pandas as pd
from sklearn.metrics import roc_curve, auc
from sklearn.metrics import accuracy_score, recall_score, precision_score
from sklearn.model_selection import cross_val_score

def displaymetrics(code: list, models: list, X_train: np.array, X_test: np.array, X: pd.DataFrame):
    for model in models:  # or for i in range(0, len(models)):
        y_score = model.fit(X_train, y_train).decision_function(X_test)
        # or y_score = models[i].fit(X_train, y_train).decision_function(X_test)
        fpr, tpr, _ = roc_curve(y_test, y_score)
        # etc etc

尝试编辑您的代码，以便向我们展示您如何调用 displaymetrics 以及使用哪些参数。

Answer 2

您应该使用字典而不是列表，如下例所示：

dict_classifiers = {
    "Logreg": LogisticRegression(solver='lbfgs'),
    "NN": KNeighborsClassifier(),
    "LinearSVM": SVC(probability=True, kernel='linear'), #class_weight='balanced'
    "GBC": GradientBoostingClassifier(),
    "DT": tree.DecisionTreeClassifier(),
    "RF": RandomForestClassifier(),
    "NB": GaussianNB(),
}

然后使用，例如：

for model, model_instantiation in dict_classifiers.iteritems():
     y_score = model_instantiation.fit(X_train, y_train).decision_function(X_test)
     ...

希望对你有帮助，让我知道你的进展如何！