recallAtN.m

function [res, recalls]= recallAtN(searcher, nQueries, isPos, ns, printN, nSample)
    if nargin<6, nSample= inf; end
    
    rngState= rng;
    
    if nSample < nQueries
        rng(43);
        toTest= randsample(nQueries, nSample);
    else
        toTest= 1:nQueries;
    end
    
    assert(issorted(ns));
    nTop= max(ns);
    
    recalls= zeros(length(toTest), length(ns));
    printRecalls= zeros(length(toTest),1);
    
    evalProg= tic;
    
    for iTestSample= 1:length(toTest)
        
        relja_progress(iTestSample, ...
                       length(toTest), ...
                       sprintf('%.4f', mean(printRecalls(1:(iTestSample-1)))), evalProg);
        
        iTest= toTest(iTestSample);
        
        ids= searcher(iTest, nTop);
        numReturned= length(ids);
        assert(numReturned<=nTop); % if your searcher returns fewer, it's your fault
        
        thisRecall= cumsum( isPos(iTest, ids) ) > 0;
        recalls(iTestSample, :)= thisRecall( min(ns, numReturned) );
        printRecalls(iTestSample)= thisRecall(printN);
    end
    t= toc(evalProg);
    
    res= mean(printRecalls);
    relja_display('\n\trec@%d= %.4f, time= %.4f s, avgTime= %.4f ms\n', printN, res, t, t*1000/length(toTest));
    
    relja_display('%03d %.4f\n', [ns(:), mean(recalls,1)']');
    
    rng(rngState);
end