《(4.3.1)--4.3SupportVectorMachine.pdf》由会员分享,可在线阅读,更多相关《(4.3.1)--4.3SupportVectorMachine.pdf(6页珍藏版)》请在得力文库 - 分享文档赚钱的网站上搜索。
1、What is Support Vector MachineSupportSupportVectorVectorMachineMachine(SVM)(SVM)isadiscriminative classifier that is formallydesigned by a separative hyperplane.It isa representation of examples as points inspace that are mapped so that the pointsof different categories are separated by agap as wide
2、 as possible.How SVM WorksExamplexWicketsyRunsWhen we plot the data,we can see a clear separation between the class of batsman&bowlerBatsmanBowlerUnknown valueWe want to classify a new player variable as a batsman or a bowlerA decision boundary is required in order to classify the new unknown variab
3、leThe decision boundary is a separation between the 2 classesWe can draw multiple lines here as decision boundariesWe cannot classify the unknown player into its correct class using multiple separating linesWe need one line that BESTseparates the dataThe best line is selected by computing the maximu
4、m margin from equidistant Support VectorsBut what exactly are Support Vectors here?Support Vectors are the points which are very close to a dividing lineUsing the Support Vectors we can select the best line to divide the dataMargin is the distance between the Support Vectors and dividing lineThe bes
5、t line will always have the greatest margin distance between the Support VectorsMargin distanceThe red line here will classify the player as a batsman.And the margin distance appears bigger than the green one.Let us consider this line as the separating boundary and classify the player variableHyperp
6、laneThis problem set is 2 Dimensional because classification is only between 2 classes2 Dimensional applications of SVM are called Linear SVMSVM KernelFeature 1Feature 2Feature 2Feature 3KernelAn SVM kernel basically adds more dimensions to a low dimensional space to make it easier to segregate the
7、data.Types of SVM KernelKernel converts the inseparable problem to separable problems byadding more dimensions using the kernel trick.Linear Kernel:be used as a normal dot product between any two givenobservationsPolynomial Kernel:generalized form of the linear kernelRadial Basis Function Kernel:can map the space in infinite dimensions