view toolboxes/distance_learning/mlr/metricPsi/metricPsiMRR.m @ 0:e9a9cd732c1e tip

first hg version after svn

function S = metricPsiMRR(q, y, n, pos, neg)
%
%   S = metricPsiMRR(q, y, n, d, pos, neg)
%
%   q   = index of the query point
%   y   = the ordering to compute
%   n   = number of points in the data set
%   pos = indices of relevant results for q
%   neg = indices of irrelevant results for q
%
%   S is the vector of weights for q

    yp          = ismember(y, pos);
    % Truncate yp after the first 1
    yp(find(yp,1)+1:end) = 0;
    NumPOS      = sum(yp);
    PosAfter    = NumPOS - cumsum(yp);

    yn          = ~yp;
    NumNEG      = sum(yn);
    NegBefore   = cumsum(yn) - yn;

    S = zeros(n,1);

    S(y)    = 2 * (yp .* NegBefore - yn .* PosAfter) / (NumNEG * NumPOS);

end