August 2018
Intermediate to advanced
522 pages
12h 45m
English
The sigmoid kernel is based on this function:

The constant term r is specified through the coef0 parameter. As pointed out in A study on sigmoid kernels for SVM and the training of non-PSD kernels by SMO-type methods, Lin, Hsuan-Tien, and Chih-Jen Lin, Submitted to Neural Computation 3 (2003): 1-32, if γ << 1 and r < 0, the sigmoid kernel behaves like an RBF one, however, in general, its performance is never dominant with respect to RBF or polynomial kernels. Hence, it's preferable to test the first two methods before trying this one.
Read now
Unlock full access