机器学习：计算训练和验证错误、不同数量的特征和应用交叉验证

Question

我试图通过改变数据点的特征数量（最多 10 个）并同时应用 5 倍交叉验证来计算训练和验证错误。我的想法是计算每次特征迭代的折叠之间的平均误差。

我找不到计算这些平均误差值的正确方法

n=10  # Number of features

err_train = np.zeros(n)    # Array for storing training errors
err_val = np.zeros(n)    # Array for storing validation errors

for i in range(n): 
    train_errors_per_cv_iteration =[] 
    test_errors_per_cv_iteration =[] 
    K=5    # Specify the number of folds of split data into
    kf = KFold(n_splits=K, shuffle=False)    # Create a KFold object with 'K' splits
    iteration = 0
    for train_indices, test_indices in kf.split(X):
        iteration += 1
        X_train = X[train_indices,:]    # Get the training set    
        X_val = X[test_indices,:]    # Get the validation set
        reg = LinearRegression(fit_intercept=True)    # create an object for linear predictors
        reg = reg.fit(X_train[:,:(i+1)], y_train)    # find best linear predictor (minimize training error)
        pred = reg.predict(X_train[:,:(i+1)])    # compute predictions of best predictors 
        err_train[i] = mean_squared_error(y_train, pred)
        pred = reg.predict(X_val[:,:(i+1)])    # compute predictions of best predictors 
        err_val[i] = mean_squared_error(y_val, pred)