K nearest neighbour predict() and knnsearch() not giving same result

1 次查看（过去 30 天）

Moritz Hesse 2019-4-24

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/458306-k-nearest-neighbour-predict-and-knnsearch-not-giving-same-result

编辑： Moritz Hesse 2019-4-24

Hi experts,

I have a ClassificationKNN object called KNNMdl which I would like to use to predict new data from my table called test_data. When I make the prediction I would also like to see the nearest neighbours used to make the prediction. However, the results of the predict and knnsearch functions yield different results

predict_row = 3176;
predicted = predict(KNNMdl, test_data{predict_row, :})
% Yields : '706'

However, when I use the knnsearch function, the highest number of closest neighbours are not from the '706' class, but from a different class:

knn_search = knnsearch(table2array(KNNMdl.X), test_data{predict_row, :},'K',20);
%knn_search returns indicies of nearest k datapoints. From this, get class labels:
nearest_classes = KNNMdl.Y(knn_search);

The nearest_classes variable shows that from the 20 closest neighbours, only 4 are in the '706' class, and the remaining 16 are in the '999' class.

What am I doing wrong? Or have I misunderstood the functionality of the knnsearch function?