Ad

Thursday, August 2, 2018

Machine Learning SVM

SVM can use other functions to make data linearly separable. SVM can give non linear, intricate decision boundaries. SVM Decision Boundary is a straight line for linear SVM.  Apply linear SVM. If it has 0% error, your data is linearly separable.

c parameter SVM controlls trade off between smooth decision boundary and classifying training points correctly (may not generalize well, get a smooth boundary or get more points classified correctly). Effects of C especially obvious in the RVF kernel. A large c means get more training points correctly. Larger c --> more intricate boundaries

Gamma Parameter
Gamma defines how far the influence of a single training example reaches. If gamma has a low value each pointer has a far reach, if gamma has a high value each point has a closer reach.  A high gamma value will make decision boundaries pay close attention to those points that are close, but ignore those that are far. High value of gamma could mean a very wiggly decision boundary.

A point close to the frontier can really have a lot of weight and pull the frontier close to itself. Versus a low gamma, means more points will have weights of influence on the frontier, so the frontier end up being smoother.

svm kernel http://scikit-learn.org/stable/modules/svm.html#svm-kernels


Use SVM for Stock finance https://en.wikipedia.org/wiki/Support_vector_machine

No comments:

Post a Comment

Applying for jobs at the Lending Club

We tried to figure out Lending Club 's tech stack for 2019. Our analysis shows Lending Club asks for skills in Python, Tableau, SQL and ...