KNN returns unexpected class (Bug #4217)

Added by Mikulas Krupicka about 10 years ago. Updated over 9 years ago.

Status:New Start date:2015-02-25
Priority:High Due date:
Assignee:Maria Dimashova % Done:

0%
Category:ml
Target version:-
Affected version:2.4.9 (latest release) Operating System:Linux
Difficulty: HW Platform:x64
Pull request:

Description

Let's assume the following code:

KNearest knn;

Mat samples = (Mat_<float>(4,1) << 1, 1.5, 2, 2.5);
Mat responses = (Mat_<int>(4,1) << 4, 3, 2, 1);
knn.train(samples, responses);

Mat test_sample = (Mat_<float>(1,1) << 1.1);
cv::Mat results, neighbours, dists;
knn.find_nearest(test_sample, 4, results, neighbours, dists);

cout << results << " - " << neighbours << ", " << dists << endl;

As the result is displayed class "1". Which is, in fact, the farthest one. I believe, the "4" should be returned instead.

Associated revisions

Revision 9ad84896
Added by Vadim Pisarevsky over 9 years ago

Merge pull request #4217 from wangyan42164:ocl_pyrlk_mul

History

#2
Updated by Nisarg Thakkar about 10 years ago

I think this is working according to the algorithm. The output is 1 because on considering k=4, we get, 1 member from each class. And so, we can choose any class as the result for that particular node. We could return the class of the nearest neighbor if there multiple classes have the same count, but that would just lead to extra computation and it is unlikely that such a situation would occur if the data set is large and distributed.

#3
Updated by Maksim Shabunin over 9 years ago

Issue has been transferred to GitHub: https://github.com/Itseez/opencv/issues/4942

Also available in: Atom PDF