scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines

scikit-learn(工程中用的相对较多的模型介绍):1.4. Support Vector Machines

参考:http://scikit-learn.org/stable/modules/svm.html


在实际项目中,我们真的很少用到那些简单的模型,比如LR、kNN、NB等,虽然经典,但在工程中确实不实用。

今天我们关注在工程中用的相对较多的SVM。


SVM功能不少:Support vector machines (SVMs) are a set of supervised learning methods used for classification, regression and outliers detection.

好处多多:高维空间的高效率;维度大于样本数的有效性;仅使用训练点的子集(称作支持向量),空间占用少;有不同的kernel functions供选择。

也有坏处:维度大于样本数的有效性----但维度如果相对样本数过高,则效果会非常差;不能直接提供概率估计,需要通过an expensive five-fold cross-validation (see Scores and probabilities, below).才能实现。

(SVM支持dense和sparse sample vectors,但是如果预测使用的sparse data,那训练也要使用稀疏数据。为了发挥SVM效用,请use C-ordered numpy.ndarray (dense) or scipy.sparse.csr_matrix (sparse) with dtype=float64.


1、分类

SVCNuSVC and LinearSVC 是三个可以进行multi-class分类的模型。三者的本质区别就是 have different mathematical formulations,具体参考本文最后的公式。

 SVCNuSVC and LinearSVC 和其他分类器一样,使用fit、predict方法:

>>> from sklearn import svm
>>> X = [[0, 0], [1, 1]]
>>> y = [0, 1]
>>> clf = svm.SVC()
>>> clf.fit(X, y)  
SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,
gamma=0.0, kernel='rbf', max_iter=-1, probability=False, random_state=None,
shrinking=True, tol=0.001, verbose=False)

After being fitted, the model can then be used to predict new values:

>>>
>>> clf.predict([[2., 2.]])
array([1])


SVM中的支持向量的相关属性可以使用 support_vectors_support_ and n_support来获取:

>>> # get support vectors
>>> clf.support_vectors_
array([[ 0.,  0.],
       [ 1.,  1.]])
>>> # get indices of support vectors
>>> clf.support_ 
array([0, 1]...)
>>> # get number of support vectors for each class
>>> clf.n_support_ 
array([1, 1]...)

对于multi-class分类:

SVC and NuSVC 的机制是“one-against-one”(training n_class * (n_class - 1) / 2个 models),而 LinearSVC 的策略是“one-vs-the-rest”(training n_class个 models) 。而实践中,one-vs-rest是常用和较好的,因为结果其实差不多,但时间省好多。。。

[python] view plaincopy
  1. >>> X = [[0], [1], [2], [3]]  
  2. >>> Y = [0123]  
  3. >>> clf = svm.SVC()  
  4. >>> clf.fit(X, Y)   
  5. SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,  
  6. gamma=0.0, kernel='rbf', max_iter=-1, probability=False, random_state=None,  
  7. shrinking=True, tol=0.001, verbose=False)  
  8. >>> dec = clf.decision_function([[1]])  
  9. >>> dec.shape[1# 4 classes: 4*3/2 = 6  
  10. 6  
  11. >>> lin_clf = svm.LinearSVC()  
  12. >>> lin_clf.fit(X, Y)   
  13. LinearSVC(C=1.0, class_weight=None, dual=True, fit_intercept=True,  
  14.      intercept_scaling=1, loss='squared_hinge', max_iter=1000,  
  15.      multi_class='ovr', penalty='l2', random_state=None, tol=0.0001,  
  16.      verbose=0)  
  17. >>> dec = lin_clf.decision_function([[1]])  
  18. >>> dec.shape[1]  
  19. 4  
关于样本所属类别的confidence:The SVC method decision_function gives per-class scores for each sample。另外还有所谓的option probability,但是,If confidence scores are required, but these do not have to be probabilities, then it is advisable to set probability=False and use decision_function instead of predict_proba.(主要是因为probability的理论背景有缺陷


在每个class或者sample的权重不同的情况下,可以设置keywords class_weight andsample_weight :

类别权重:SVC (but not NuSVC) implement a keyword class_weight in the fit method. It’s a dictionary of the form {class_label : value}, where value is a floating point number > 0 that sets the parameter C of class class_label to C * value.

样本权重:SVCNuSVCSVRNuSVR and OneClassSVM implement also weights for individual samples in method fit through keyword sample_weight. Similar to class_weight, these set the parameter C for the i-th example to C * sample_weight[i].


最后给几个例子:

  • Plot different SVM classifiers in the iris dataset,
  • SVM: Maximum margin separating hyperplane,
  • SVM: Separating hyperplane for unbalanced classes
  • SVM-Anova: SVM with univariate feature selection,
  • Non-linear SVM
  • SVM: Weighted samples,


2、回归

Support Vector Regression.

看能明白这句话不能:Analogously(to SVClassfication), the model produced by Support Vector Regression depends only on a subset of the training data, because the cost function for building the model ignores any training data close to the model prediction.

同样也是三个模型: SVRNuSVR and LinearSVR

>>> from sklearn import svm
>>> X = [[0, 0], [2, 2]]
>>> y = [0.5, 2.5]
>>> clf = svm.SVR()
>>> clf.fit(X, y) 
SVR(C=1.0, cache_size=200, coef0=0.0, degree=3, epsilon=0.1, gamma=0.0,
    kernel='rbf', max_iter=-1, shrinking=True, tol=0.001, verbose=False)
>>> clf.predict([[1, 1]])
array([ 1.5])
给个例子:

  • Support Vector Regression (SVR) using linear and non-linear kernels

3、Density estimation,novelty detection(密度估计、新颖性检测)

先看下wiki上怎么说Novelty detection:Novelty detection is the identification of new or unknown data that a machine learning system has not been trained with and was not previously aware of,[1] with the help of either statistical or machine learning based approaches.

 OneClassSVM is used for novelty detection, that is, given a set of samples, it will detect the soft boundary of that set so as to classify new points as belonging to that set or not. 过程是无监督的,所以输入只有X。

具体详细应用参考:section Novelty and Outlier Detection 。

最后给出两个例子:

  • One-class SVM with non-linear kernel (RBF)
  • Species distribution modeling


4、复杂度

The QP(quadratic programming problem) solver used by this libsvm-based implementation scales between scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines and scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines depending on how efficiently the libsvm cache is used in practice (dataset dependent).



5、实际应用中的一些小tips

Avoid data copy;kernel cache size;

Setting C:C默认是1,但是如果data中有很多noisy observations,需要减小C;

it is highly recommended to scale your data. For example, scale each attribute on the input vector X to [0,1] or [-1,+1], or standardize it to have mean 0 and variance 1. Note that the same scaling must be applied to the test vector to obtain meaningful results. 

在 SVC中,如果数据样本unbalanced,set class_weight='auto' and/or try different penalty parameters C.


6、kernel function

使用方式为:svm.SVC(kernel='linear'),常见的kernel有:

  • linear: scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines.
  • polynomial: scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machinesscikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines is specified by keyword degreescikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines by coef0.
  • rbf: scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machinesscikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines is specified by keyword gamma, must be greater than 0.
  • sigmoid (scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines), where scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines is specified by coef0.
也可自定义kernel,例如:

>>> import numpy as np
>>> from sklearn import svm
>>> def my_kernel(x, y):
...     return np.dot(x, y.T)
...
>>> clf = svm.SVC(kernel=my_kernel)
  • SVM with custom kernel.

7、Mathematical formulation

1、SVC:

scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines

2、SVR:

scikit-learn(工程顶用的相对较多的模型介绍):1.4. Support Vector Machines


版权声明:本文为博主原创文章,未经博主允许不得转载。