Practical notes for machine learning 6 (SVM)

In view of July Big Brother SVM three level realm (http://blog.csdn.net/v_july_v/article/details/7624837) has been written very well, here I will not detail, just elaborate on a few simple concepts. If I am confused about the layer-3 realm of SVM, I am willing to communicate with you and make progress together.

Simple concept description:

(1) Support Vector Machine (SVM) is to maximize the classification interval between the support vector and the classification hyperplane. The classification hyperplane is the decision surface we want to obtain. The support vector is the point closest to the classification hyperplane, and the interval is the distance from the support vector to the classification hyperplane.

(2) kernel functions: Generally, the function of kernel functions is to map data from a low-dimensional space to a high-dimensional space, and linear division is not possible. The meaning of this sentence is illustrated in a simple example: A1 * X1 ^ 2 + A2 * X2 ^ 2 + A3 * x1x2 = 0, in this case, z1 = x1 ^ 2, Z2 = x2 ^ 2, Z3 = x1x2, Which is mapped from the original two-dimensional space to the three-dimensional space. At this time, it also becomes linearly segmented, the corresponding ing function is represented by P. The function of kernel functions is to calculate the inner product <p (X1) and P (X2)> when solving SVM. However, in a high-dimensional space, the calculation of inner product is often complicated, sometimes there may be a dimension disaster, so we can use the kernel function to solve this problem. If you still can't understand it, you can refer to the example of the second-level kernel function in July's blog.

Note: (1) SVM is the best classifier that can be used without any modifications.

(2) SVM has many implementation methods. The most common method is the sequence least optimization algorithm (SMO, sequentialminimal optimization)

(3) SVM can be used for almost all classification problems, but it is worth mentioning that SVM itself is a binary classifier, to apply SVM to multiclass classification problems, you need to modify the code.

(4) SVM is a classifier. It is called "machine" because it produces a binary decision result, that is, it is a decision "machine ".