为什么函数会以不同的格式和顺序返回结果

Question

这段代码运行良好，令人期待。拟合、预测和计算损失函数：

features = set(df_train_cat.columns)
features.remove('SalePrice')
features.remove('Id')
X_train_cat = df_train_cat[features].values
y_train_cat = df_train_cat.SalePrice.values
X_train, X_val, y_train, y_val = train_test_split( X_train_cat, y_train_cat, test_size=0.33,                         random_state=17)

regression = linear_model.Ridge(alpha=.5, random_state = 17)
regression.fit(X_train,y_train)
predictions = regression.predict(X_val)
rmse = (mean_squared_error(y_val, predictions))**(1/2)
mae = mean_absolute_error(y_val, predictions)
print('RMSE on run = {}, MAE = {}'.format(rmse, mae))
-------------------------------------------------------
in returns: 
RMSE on run = 27696.648368899852, MAE = 17553.925337213957

但当我试图将其封装在函数中时，结果却不同：

def regress_without_uncorrelated(delete_feture_list, data):
features = set(data.columns)
delete_feture_list = delete_feture_list + ['SalePrice', 'Id']
if len(delete_feture_list) > 2:
features.remove(set(delete_feture_list))

X = data[features].values
y = data.SalePrice.values
X_train, X_val, y_train, y_val = train_test_split( X, y, test_size=0.33, random_state=17)
regression = linear_model.Ridge(alpha=.5, random_state = 17)
regression.fit(X_train,y_train)
predictions = regression.predict(X_val)

rmse = (mean_squared_error(y_val, predictions))**(1/2)
mae = mean_absolute_error(y_val, predictions)
print('rmse = {}, mae = {}'.format(rmse,  mae))
return rmse, mae
-------------------------------------
result is:         
rmse = 2.760396036595275e-08, mae = 1.7484699770097167e-08

问题：

1. 为什么结果返回e？如何让它像第一个街区一样
2. 为什么在固定random_state的情况下，相同数据的丢失顺序会发生变化

Answer 1

问题出现在set.remove((方法中，该方法不适用于iterable。