线形回归-sklearn

对于先行回归的计算，有两种方法。一种是直接计算，一种利用了梯度下降算法。

梯度下降，相当于一元函数里的倒数的概念。在高维的条件下，通过求偏导来决定下降的最快路径。

API

from sklearn.linear_model import LinearRegression, SGDRegressor
from sklearn.datasets import load_boston
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import mean_squared_error


boston = load_boston()
x_train, x_test, y_train, y_test = train_test_split(
    boston.data, boston.target, random_state=10)

trans = StandardScaler()
x_train = trans.fit_transform(x_train)
x_test = trans.transform(x_test)

# 直接计算，正规方程
estimator = LinearRegression()
estimator.fit(x_train, y_train)
pre = estimator.predict(x_test)
error = mean_squared_error(y_test, pre)
print(error)

# 随机梯度下降
estimator = SGDRegressor(
    random_state=0, learning_rate='invscaling', eta0=0.01, max_iter=10000)
# learn_rate:
# constant:学习过程中的学习率不变
# optimal : 1.0 / (alpha * (t + t0))
# invscaling : eta = eta0 / pow(t, power_t) power_t = 0.25
# eta0 可以指定初始学习率
estimator.fit(x_train, y_train)
pre = estimator.predict(x_test)
error = mean_squared_error(y_test, pre)
print(error)

32.4425366960067
32.50484444073032

随机梯度下降在神经网络里有十分重要的作用。