尝试适应 GridSearchCV 时出错

Question

我正在尝试使用管道和 GridSearchCV 将岭回归模型拟合到我的数据中。

from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import PolynomialFeatures
from sklearn.pipeline import Pipeline

X = transformed_data.iloc[:, :-1]
y = transformed_data['class'] 

X_train, X_test, y_train, y_test = train_test_split(X, y, random_state = 1)

params = {}
params['ridge__alpha'] = np.arange(0, 100, 1).tolist()

t = [('labelenc',LabelEncoder() , [0]), ('stand', StandardScaler(), [1,2,3,4,5,6]), ('poly'),PolynomialFeatures(degree=2),[1,2,3,4,5,6] ]
transformer = ColumnTransformer(transformers=t)


pipe = Pipeline(steps=[('t', transformer), ('m',Ridge())])


#grid_ridge2_r2 = GridSearchCV(pipe, params, cv=10, scoring='r2', n_jobs=-1)
#results_ridge2_r2 = grid_ridge2_r2.fit(X_train,y_train)

grid_ridge2_rmse = GridSearchCV(pipe, params, cv=10, scoring='neg_root_mean_squared_error', n_jobs=-1)
results_ridge2_rmse = grid_ridge2_rmse.fit(X_train,y_train)

我不断得到

ValueError: too many values to unpack (expected 3)

在最后一行 grid_ridge2_rmse.fit(X_train,y_train)。我的直觉是我拆分数据集的方式有问题。

Answer 1

您的管道中存在一些错误。

首先 LabelEncoder 不能在 scikit-learn 管道中使用，因为它用于修改 y 而不是 X。假设您要对特征的分类值进行编码，则应将其替换为 OrdinalEncoder.

然后，要设置网格参数，它必须按照以下命名约定命名 <step>__<hyperparameter。在你的情况下设置山脊参数应该是 m__alpha.

管道参数可以使用pipe.get_params()查看。

我会这样做：

from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import PolynomialFeatures, OrdinalEncoder, StandardScaler
from sklearn.linear_model import Ridge
from sklearn.model_selection import GridSearchCV
from sklearn.pipeline import Pipeline
import numpy as np

X_train, X_test, y_train, y_test = train_test_split(X, y, random_state = 1)

params = {'m__alpha' : np.arange(0, 100, 1).tolist()}

t = [
    ('labelenc',OrdinalEncoder() , [0]),
    ('stand', StandardScaler(), [1,2,3,4,5,6]),
    ('poly', PolynomialFeatures(degree=2), [1,2,3,4,5,6])
]

transformer = ColumnTransformer(transformers=t)


pipe = Pipeline(steps=[('t', transformer), ('m',Ridge())])

grid_ridge2_rmse = GridSearchCV(pipe, params, cv=10, scoring='neg_root_mean_squared_error', n_jobs=-1)
results_ridge2_rmse = grid_ridge2_rmse.fit(X_train,y_train)

尝试适应 GridSearchCV 时出错

Getting error while trying to fit to GridSearchCV

machine-learning

scikit-learn

gridsearchcv