Bayes+linearSVM, Bayes+k-NN
The worker here has the same average (over classes) reliability.
The results show that k-NN sometimes beat linearSVM, especially when the risk threshold is low (=0.)
Created on July 10|Last edited on July 13
Comment
Linear SVM vs. k-NN
Different risk threshold
k-NN method is more resistant to low risk threshold
Risk threshold=0.
72
0
0
Different k
Large k performs better
k=5
361
0
0
Note that the workers here are all reliable to some extent.
What happened to linearSVM? (when risk threshold=0.)
Run set
5520
Overfit to the noisy label (# features=8092,# data =10*50=500)
Add a comment