如何用Sklearn高效实现LASSO模型的固定验证集调参？

我需要用sklearn训练LASSO模型，已有专门划分好的训练集和验证集。目标是自动生成一系列alpha值（L1惩罚强度），针对每个alpha用训练集拟合模型后在验证集评估性能，最终选择验证集表现最优的模型。

已尝试的方法 #

1. LassoCV + 自定义1折CV迭代器 #

将训练集和验证集绑定为1折交叉验证迭代器传入cv参数，但fit()方法最终会用最优alpha和合并后的全量数据训练模型，必须手动取出最优alpha再重新调用Lasso训练，步骤繁琐：

import numpy as np
from sklearn.linear_model import LassoCV
from sklearn.datasets import make_regression
X, y = make_regression(noise = 4, random_state = 0)
Nrow, Ncol = len(X), len(X[0])
Ntrain = int(np.round(Nrow * 0.7))
Nvalid = Nrow - Ntrain
trainInd = np.asarray([i for i in range(Ntrain)])
validInd = np.asarray([i for i in range(Ntrain, Nrow)])
trainValidInd = [(trainInd, validInd)]
cvIter = iter(trainValidInd)

reg = LassoCV(cv = cvIter, verbose = True).fit(X, y) 
# 问题：fit()会用最优alpha和合并后的全部数据训练模型

2. lasso_path #

尝试过sklearn.linear_model.lasso_path，但不清楚如何结合验证集做预测，且该方法不返回截距项，无法获取模型截距。

3. 临时 workaround #

给验证集样本设置极小权重（接近0），让模型拟合时忽略验证集，测试有效但方法不够合理：

sampleW = np.asarray([1.0 for i in range(Ntrain)] + \
    [1e-200 for i in range(Nvalid)]) 
reg = LassoCV(cv = cvIter, verbose = True).fit(X, y, sampleW)

合理高效的实现方案 #

方案1：手动遍历候选alpha #

直接拆分训练/验证集，用LassoCV生成候选alpha（或手动指定），逐个训练并评估，最终选择最优模型：

import numpy as np
from sklearn.linear_model import Lasso, LassoCV
from sklearn.datasets import make_regression
from sklearn.metrics import mean_squared_error

# 拆分训练集和验证集
X, y = make_regression(noise=4, random_state=0)
Ntrain = int(np.round(len(X) * 0.7))
X_train, X_valid = X[:Ntrain], X[Ntrain:]
y_train, y_valid = y[:Ntrain], y[Ntrain:]

# 生成候选alpha（基于训练集的LassoCV自动生成）
lasso_cv = LassoCV(cv=5, random_state=0)
lasso_cv.fit(X_train, y_train)
candidate_alphas = lasso_cv.alphas_

# 或者手动指定alpha范围
# candidate_alphas = np.logspace(-4, 4, 50)

# 遍历每个alpha，训练并评估
best_score = float('inf')
best_alpha = None
best_model = None

for alpha in candidate_alphas:
    model = Lasso(alpha=alpha, random_state=0)
    model.fit(X_train, y_train)
    y_pred = model.predict(X_valid)
    # 用MSE作为评估指标，可替换为其他指标如R²
    score = mean_squared_error(y_valid, y_pred)
    if score < best_score:
        best_score = score
        best_alpha = alpha
        best_model = model

print(f"最优alpha: {best_alpha:.4f}")
print(f"验证集最优MSE: {best_score:.4f}")

方案2：GridSearchCV + 自定义CV迭代器 #

利用GridSearchCV，指定自定义1折CV迭代器，同时设置refit=False避免用全量数据重训，最后用最优alpha训练模型：

import numpy as np
from sklearn.linear_model import Lasso
from sklearn.datasets import make_regression
from sklearn.model_selection import GridSearchCV

# 拆分数据并生成CV迭代器
X, y = make_regression(noise=4, random_state=0)
Ntrain = int(np.round(len(X) * 0.7))
trainInd = np.asarray([i for i in range(Ntrain)])
validInd = np.asarray([i for i in range(Ntrain, len(X))])
trainValidInd = [(trainInd, validInd)]

# 定义参数网格
candidate_alphas = np.logspace(-4, 4, 50)
param_grid = {'alpha': candidate_alphas}

# 初始化GridSearchCV，refit=False避免全量重训
grid_search = GridSearchCV(
    estimator=Lasso(random_state=0),
    param_grid=param_grid,
    cv=trainValidInd,
    refit=False,
    scoring='neg_mean_squared_error'  # 负MSE，GridSearch默认最大化得分
)
grid_search.fit(X, y)

# 获取最优参数和得分
best_alpha = grid_search.best_params_['alpha']
best_score = -grid_search.best_score_  # 转为正MSE

# 用最优alpha训练最终模型
best_model = Lasso(alpha=best_alpha, random_state=0)
best_model.fit(X[:Ntrain], y[:Ntrain])

print(f"最优alpha: {best_alpha:.4f}")
print(f"验证集最优MSE: {best_score:.4f}")

这两种方案都清晰高效，避免了临时 workaround 的不合理性，同时严格遵循训练/验证集的划分逻辑。

内容的提问来源于stack exchange，提问作者user2961927