Hello, In k-nn, in order to compute the training error (given our training set) do we always assume that a point have to be itself one of the k nearest neighbours (thus the closest one) ? Thanks.
Yes other wise you wouldnt be able to get a training error = 0 when k = 1
neighbours in knn
Hello,
In k-nn, in order to compute the training error (given our training set) do we always assume that a point have to be itself one of the k nearest neighbours (thus the closest one) ?
Thanks.
1
Yes other wise you wouldnt be able to get a training error = 0 when k = 1
Add comment