Can recognize what distance is this ?

Question

Kamil Kacer 2020-11-13

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/646788-can-recognize-what-distance-is-this

评论： Kamil Kacer 2020-11-13

Can anyone explain me what does this code do,

Could it be the after first if is calculated euclodean distance and in else there is second distance called mahalanobis distance ?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Walter Roberson 2020-11-13

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/646788-can-recognize-what-distance-is-this#answer_543543

The clue is in the name of the variable: useL1Distance . If it is true, then the L1 norm is used, which is also called "taxi-cab distance".

If it is false, then the L2 norm is used, which is also called Euclidean distance.

Neither branch has anything to do with maholonobis distance.

4 个评论
显示 2更早的评论隐藏 2更早的评论

Walter Roberson 2020-11-13

在 MATLAB Online 中打开

d{i} = pdist2(F{i}.', testSample, 'mahalanobis');

Kamil Kacer 2020-11-13

Thank you it worked iam assuming it calculates the distance between testsample and every audio segment F in the dataset a stores it into d

Iam I right ?

请先登录，再进行评论。

Answer 2

Kamil Kacer 2020-11-13

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/646788-can-recognize-what-distance-is-this#answer_543583

在 MATLAB Online 中打开

I have this function where iam clyssifing a sample and i just want to another distance based on which is the sample classified

I have euclidean and I want to add mahalanobis as you showed me. but it doesnt work any suggestions please.

function [Ps, winnerClass] = classifyKNN_D_Multi(F, testSample, k, NORMALIZE, useL1distance )
% function [Ps, winnerClass] = classifyKNN_D_Multi(F, testSample, k, NORMALIZE, useL1distance);
% 
% This function is used for classifying an unknown sample using the kNN
% algorithm, in its multi-class form.
%
% ARGUMENTS:
% - F: an CELL array that contains the feature values for each class. I.e.,
%      F{1} is a matrix of size numOfDimensions x numofSamples FOR THE FIRST
%      CLASS, etc.
%
% - testSample: the input sample to be classified
% - k: the kNN parameter
% - NORMALIZE: use class priors to weight results
% - useL1distance: use L1 instead of L2 distance
%
% RETURNS:
% - Ps: an array that contains the classification probabilities for each class
% - winnerClass: the label of the winner class
error( nargchk(4,5,nargin) )
if ( nargin < 5 )
    useL1distance = false;
end
numOfClasses = length(F);
if (size(testSample, 2)==1)
    testSample = testSample';
end
% initilization of distance vectors:
numOfDims = zeros( 1, numOfClasses );
numOfTrainSamples = zeros( 1, numOfClasses );
d = cell(numOfClasses,1);
% d{i} is a vector, whose elements represent the distance of the testing
% sample from all the samples of i-th class
testSample(isnan(testSample)) = 0.0;
for i=1:numOfClasses
    [ numOfDims(i), numOfTrainSamples(i) ] = size( F{i} );
    d{i} = inf*ones(max(numOfTrainSamples), 1); % we fill it with inf values
    F{i}(isnan(F{i})) = 0.0;    
end
if (length(testSample)>1)
    for i=1:numOfClasses % for each class:
        if (numOfTrainSamples(i)>0)
            if ( useL1distance )                
                d{i} = sum( abs(repmat(testSample, [numOfTrainSamples(i) 1]) - F{i}'),2); % L1
            else
                %[size(repmat(testSample, [numOfTrainSamples(i) 1])) size(F{i}')]
                %sum(sum(isnan(F{i}')))
              %  d{i} = sum( ((repmat(testSample, [numOfTrainSamples(i) 1]) - F{i}').^2 ),2); % L2 
                 d{i} =  pdist(testSample, F{i}, 'mahalanobis');
            end
            d{i} = sort(d{i});            
            d{i}(end+1:max(numOfTrainSamples)) = inf;                                    
        else
            d{i} = inf;
        end        
    end
else % single dimension (NO SUM required!!!)
    for i=1:numOfClasses
        if (numOfTrainSamples(i)>0)
            d{i}   = (abs(repmat(testSample, [numOfTrainSamples(i) 1]) - F{i}')'); 
            d{i} = sort(d{i});
            d{i}(end+1:max(numOfTrainSamples)) = inf;            
        else
            d{i} = inf;
        end
    end    
end
kAll = zeros(numOfClasses, 1);
for j=1:k
    curArray = zeros(numOfClasses, 1);
    for i=1:numOfClasses
        curArray(i) = d{i}(kAll(i)+1);
    end    
    [MIN, IMIN] = min(curArray);
    kAll(IMIN) = kAll(IMIN) + 1;
end
if ( NORMALIZE == 0 )
    Ps = (kAll ./ k);
else
    Ps = kAll ./ numOfTrainSamples';
    Ps = Ps / sum(Ps);
end
[MAX, IMAX] = max(Ps);
winnerClass = IMAX;