岭回归和普通最小二乘法回归的一个重要区别是前者对系数模的平方进行了限制。最小化目标函数,如下所示:
其中,参数控制着收缩程度,其值越大,收缩程度越大。如下所示:
In [1]: from sklearn import linear_model In [2]: clf = linear_model.R linear_model.RandomizedLasso linear_model.RandomizedLogisticRegression linear_model.Ridge linear_model.RidgeCV linear_model.RidgeClassifier linear_model.RidgeClassifierCV In [2]: clf = linear_model.Ridge(alpha = .5) In [3]: clf.fit([[0, 0], [0, 0], [1, 1]], [0, .1, 1]) Out[3]: Ridge(alpha=0.5, copy_X=True, fit_intercept=True, max_iter=None, normalize=False, solver='auto', tol=0.001) In [4]: clf.coef_ Out[4]: array([ 0.34545455, 0.34545455]) In [5]: clf.intercept_ Out[5]: 0.13636363636363641解析:
(1)sklearn.linear_model.Ridge类构造方法
class sklearn.linear_model.Ridge(alpha=1.0, fit_intercept=True, normalize=False, copy_X=True, max_iter=None, tol=0.001, solver='auto')
(2)sklearn.linear_model.Ridge类实例的属性和方法
(3)Ridge Regression(岭回归)岭回归分析是一种专用于共线性数据分析的有偏估计回归方法,实质上是一种改良的最小二乘估计法,通过放弃最小二乘法的无偏性,以损失部分信息、降低精度为代价获得回归系数更为符合实际、更可靠的回归方法,对病态数据的耐受性远远强于最小二乘法。岭回归分析主要解决两类问题:数据点少于变量个数;变量间存在共线性。
Plot Ridge coefficients as a function of the regularization
print(__doc__) import numpy as np import pylab as pl from sklearn import linear_model # X is the 10x10 Hilbert matrix X = 1. / (np.arange(1, 11) + np.arange(0, 10)[:, np.newaxis]) y = np.ones(10) ############################################################################### # Compute paths n_alphas = 200 alphas = np.logspace(-10, -2, n_alphas) clf = linear_model.Ridge(fit_intercept=False) coefs = [] for a in alphas: clf.set_params(alpha=a) clf.fit(X, y) coefs.append(clf.coef_) ############################################################################### # Display results ax = pl.gca() ax.set_color_cycle(['b', 'r', 'g', 'c', 'k', 'y', 'm']) ax.plot(alphas, coefs) ax.set_xscale('log') ax.set_xlim(ax.get_xlim()[::-1]) # reverse axis pl.xlabel('alpha') pl.ylabel('weights') pl.title('Ridge coefficients as a function of the regularization') pl.axis('tight') pl.show()图形输出,如下所示:
解析:
(1)希尔伯特矩阵
在线性代数中,希尔伯特矩阵是一种系数都是单位分数的方块矩阵。具体来说一个希尔伯特矩阵H的第i横行第j纵列的系数是:
举例来说,的希尔伯特矩阵就是:
希尔伯特矩阵的系数也可以看作是以下积分:
也就是当向量是关于变量x 的各阶幂时关于积分范数的格拉姆矩阵。
希尔伯特矩阵是低条件矩阵的典型例子。与希尔伯特矩阵的数值计算是十分困难的。举例来说,当范数为矩阵范数时希尔伯特矩阵的条件数大约是,远大于1。
(2)np.arange()方法
In [31]: 1. / (np.arange(1, 11)) Out[31]: array([ 1. , 0.5 , 0.33333333, 0.25 , 0.2 , 0.16666667, 0.14285714, 0.125 , 0.11111111, 0.1 ]) In [32]: (1. / (np.arange(1, 11))).shape Out[32]: (10,)
(3)np.newaxis属性
In [5]: np.arange(0, 10) Out[5]: array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9]) In [6]: type(np.arange(0, 10)) Out[6]: numpy.ndarray In [7]: np.arange(0, 10).shape Out[7]: (10,) In [8]: np.arange(0, 10)[:, np.newaxis] Out[8]: array([[0], [1], [2], [3], [4], [5], [6], [7], [8], [9]]) In [9]: np.arange(0, 10)[:, np.newaxis].shape Out[9]: (10, 1)(4)广播原理
In [25]: x = np.arange(0, 5) In [26]: x[:, np.newaxis] Out[26]: array([[0], [1], [2], [3], [4]]) In [27]: x[np.newaxis, :] Out[27]: array([[0, 1, 2, 3, 4]]) In [28]: x[:, np.newaxis] + x[np.newaxis, :] Out[28]: array([[0, 1, 2, 3, 4], [1, 2, 3, 4, 5], [2, 3, 4, 5, 6], [3, 4, 5, 6, 7], [4, 5, 6, 7, 8]])(5)10阶希尔伯特矩阵X
In [33]: X = 1. / (np.arange(1, 11) + np.arange(0, 10)[:, np.newaxis]) In [34]: X Out[34]: array([[ 1. , 0.5 , 0.33333333, 0.25 , 0.2 , 0.16666667, 0.14285714, 0.125 , 0.11111111, 0.1 ], [ 0.5 , 0.33333333, 0.25 , 0.2 , 0.16666667, 0.14285714, 0.125 , 0.11111111, 0.1 , 0.09090909], [ 0.33333333, 0.25 , 0.2 , 0.16666667, 0.14285714, 0.125 , 0.11111111, 0.1 , 0.09090909, 0.08333333], [ 0.25 , 0.2 , 0.16666667, 0.14285714, 0.125 , 0.11111111, 0.1 , 0.09090909, 0.08333333, 0.07692308], [ 0.2 , 0.16666667, 0.14285714, 0.125 , 0.11111111, 0.1 , 0.09090909, 0.08333333, 0.07692308, 0.07142857], [ 0.16666667, 0.14285714, 0.125 , 0.11111111, 0.1 , 0.09090909, 0.08333333, 0.07692308, 0.07142857, 0.06666667], [ 0.14285714, 0.125 , 0.11111111, 0.1 , 0.09090909, 0.08333333, 0.07692308, 0.07142857, 0.06666667, 0.0625 ], [ 0.125 , 0.11111111, 0.1 , 0.09090909, 0.08333333, 0.07692308, 0.07142857, 0.06666667, 0.0625 , 0.05882353], [ 0.11111111, 0.1 , 0.09090909, 0.08333333, 0.07692308, 0.07142857, 0.06666667, 0.0625 , 0.05882353, 0.05555556], [ 0.1 , 0.09090909, 0.08333333, 0.07692308, 0.07142857, 0.06666667, 0.0625 , 0.05882353, 0.05555556, 0.05263158]])
(6)np.ones()方法
In [35]: y = np.ones(10) In [36]: y Out[36]: array([ 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]) In [37]: y.shape Out[37]: (10,)
(7)numpy.logspace()方法
numpy.logspace(start, stop, num=50, endpoint=True, base=10.0)
说明:
Return numbers spaced evenly on a log scale. In linear space, the sequence starts at base ** start (base to the power of start) and ends with base ** stop (see endpoint below).
In [38]: n_alphas = 200 In [39]: alphas = np.logspace(-10, -2, n_alphas) In [40]: alphas Out[40]: array([ 1.00000000e-10, 1.09698580e-10, 1.20337784e-10, 1.32008840e-10, 1.44811823e-10, 1.58856513e-10, 1.74263339e-10, 1.91164408e-10, 2.09704640e-10, ..., 5.23109931e-03, 5.73844165e-03, 6.29498899e-03, 6.90551352e-03, 7.57525026e-03, 8.30994195e-03, 9.11588830e-03, 1.00000000e-02]) In [41]: alphas.shape Out[41]: (200,) In [42]: 1.00000000e-10 Out[42]: 1e-10(8)set_params(**params)方法
(9)matplotlib.pyplot.gca(**kwargs)方法
Return the current axis instance. This can be used to control axis properties either using set or the Axes methods, for example, setting the x axis range.
参考文献:
[1] 岭回归: http://baike.baidu.com/link?url=S1DwT9XFOthlB5hjGP6Ramxt-fvtCJ-RUXYVSw-z9t7-hZIojL7eroUQwKaJd5KE9-jVEQeRtxZeuUz59SBE6q
[2] 正则化、归一化含义解析: http://sobuhu.com/ml/2012/12/29/normalization-regularization.html
[3] 希尔伯特矩阵: http://zh.wikipedia.org/zh-cn/%E5%B8%8C%E5%B0%94%E4%BC%AF%E7%89%B9%E7%9F%A9%E9%98%B5
[4] 岭回归分析总结: http://download.csdn.net/detail/shengshengwang/7225251