Why kNNsearch and kNNclassify don't give the same result??

I'm using knnclassify to do a kNN classification in Matlab and it's working well. Now, I need to know the distances to the neighbors and it seems that knnsearch funcion gives me that.

The problem is the results are not the same. I am quite sure knnsearch is not working properly, but I don't know the reason.

This is my code:

k=1;
distance='euclidean';
rule='nearest';

%kNN classification
result = knnclassify(sample_matrix, training_matrix, label_matrix,k,distance,rule);

%showing which element is recognized and the distance to it
[recognition,distances] = knnsearch(training_matrix, sample_matrix,'k',k);
So, result and recognition should be the same, and then I can see the distances.

This is result:

4 1 1 2 4 1 1 4 1 1 2 2 2 2 2 2 2 2 2 2 3 3 3 2 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 5 5 5 5 5 5 5 5 5 5 5 6 6 6 6 6 6 6 6 6 7 7 7 7 7 7 7 7 7 7 8 8 8 8 8 8 8 8 8 8 9 9 9 9 9 9 9 9 9 9

And this is recognition:

3 1 1 2 3 1 1 3 1 1 2 2 1 2 2 2 1 2 2 2 2 2 2 2 2 2 2 3 2 2 4 3 3 4 3 4 4 3 3 3 4 4 4 4 4 4 4 4 4 4 4 5 5 5 5 5 5 5 5 5 5 5 6 6 6 5 5 6 6 6 7 6 6 6 6 7 6 6 6 7 8 7 7 7 7 7 7 7 7 7

(They're two vectors). The desired result is

1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 5 5 5 5 5 5 5 5 5 5 6 6 6 6 6 6 6 6 6 6 7 7 7 7 7 7 7 7 7 7 8 8 8 8 8 8 8 8 8 8 9 9 9 9 9 9 9 9 9 9

So, result is almost well (good enough) and recognition is a disaster. As you can see the beginning is much better than the end.

I don't understand because a simple example does work:

sample = [.9 .8;.1 .3;.2 .6];  training=[0 0;.5 .5;1 1];  group = [1;2;3];  foo_classify = knnclassify(sample, training, group); [foo_search,dist] = knnsearch(training, sample);  The results are:  foo_search = 3 1 2; foo_classify = 3 1 2; dist = 0.2236 0.3162 0.3162;

Anybody can help me?? Thank you very much.

Answer 1

[Tom Lane]

As I understand this, knnclassify returns a classification so result is a value from label_matrix.

knnsearch just searches for nearest neighbors with no classification involved, so recognition would be a row number to index into training_matrix.

If label_matrix is just a row number, then maybe that's okay. In that case I'd check to see if the nearest neighbor is a tie, and the two functions are just choosing two different equidistant neighbors.