我试图使用手动代码复制Sklearn线性回归库中的成本结果。两者之间存在巨大差异，我无法找出原因。这是Sklearn的代码:
SkLearn实现:
X_train, X_test, Y_train, Y_test = model_selection.train_test_split(X, Y, test_size=0.30)classifier = sklearn.linear_model.LinearRegression()classifier.fit(X_train,Y_train)cost = np.sqrt(np.sum((np.dot(X_train,classifier.coef_.reshape(9,1)) + classifier.intercept_ - Y_train.reshape(478,1))**2))print(cost)cost = 4.236441942240197我尝试复制的代码:
a = X_train_rev.shapeassert(X_train_rev.shape == (478,10)) # 断言X_train_rev的形状Y_train = Y_train.reshape(478,1)alpha = 0.0005 # 学习率coefficient = np.random.randn(1,10) # 初始化系数，包括截距# 循环迭代for i in range(100000): cost = np.sqrt(np.sum((np.dot(X_train_rev,coefficient.T) - Y_train)**2)) # 成本结果 if i % 10000 == 0: print(cost) grad = np.dot((np.dot(X_train_rev,coefficient.T) - Y_train).T, X_train_rev) # 计算梯度 coefficient = coefficient - (alpha * grad) # 调整系数，包括截距迭代后的成本:45.2304286497357910.42840191696328510.42840191696328510.42840191696328510.42840191696328510.42840191696328510.42840191696328510.42840191696328510.42840191696328510.428401916963285根据我的手动代码，成本未能进一步降低，并且与Sklearn的成本相差甚远。我尝试调整alpha变量，但任何增加alpha都会导致成本趋向于正无穷大。
请注意，我的手动代码中使用的X_train_rev数据有10列/特征，而不是Sklearn训练集中9个特征，因为我在训练集中添加了一列’ones’来表示截距。同样，系数向量也包括截距。
回答：
我已经尝试复制您的问题
from sklearn.datasets import make_regressionfrom sklearn.model_selection import train_test_splitfrom sklearn.linear_model import LinearRegressionfrom sklearn.metrics import mean_squared_errorX, Y = make_regression(n_samples=500, n_features=9, bias=0, random_state=1)X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.30, random_state=1)classifier = LinearRegression(fit_intercept=False)classifier.fit(X_train,Y_train)cost = np.sqrt(np.sum((np.dot(X_train,classifier.coef_.reshape(9,1)) + classifier.intercept_ - Y_train.reshape(-1,1))**2))print(cost)print('Manual regression')Y_train = Y_train.reshape(-1,1)alpha = 0.0005 # 学习率coefficient = np.random.randn(1,9) # 初始化系数，包括截距# 循环迭代for i in range(100000): cost = np.sqrt(np.sum((np.dot(X_train,coefficient.T) - Y_train)**2)) # 成本结果 if i % 10000 == 0: print(cost) grad = np.dot((np.dot(X_train,coefficient.T) - Y_train).T, X_train) # 计算梯度 coefficient = coefficient - (alpha * grad) # 调整系数，包括截距做了一些小的调整以使代码完全可重现。我没有遇到任何问题。MSE之间存在微小差异，但两者的得分都小于1e-11，所以这是一个数值问题。
