hard20 pts · first answer countsmachine-learningsvmscalability

Kernel SVM on ten million rows

Training an RBF-kernel SVC on 10 million rows never finishes. Why, and what is a common alternative?

A.SVC supports at most one million features; drop some featuresB.RBF kernels need integer inputs; round the data before trainingC.SVC trains only on a GPU; move the job to GPU hardwareD.Cost grows at least as n²; use a linear model or kernel approximation

Discussion 0

Join free to post in the discussion.

Kernel SVM on ten million rows

Discussion 0

Related questions