Nearest-Neighbour-Search-Neural-Networks-Product-Quantization/TEXMEX/Plot_Run_WNNetsOverPQ_Fine_Coarse.m

clc; clear all; close all

kkf=256;
mmf=16;

nn=100;
L0=1000;
mmc=2.^[1:4];
kkc=[1:9, 10:10:100, 200:100:1000];

lnn=length(nn); lmm=length(mmc); lkk=length(kkc);

for inn=1:lnn;
    figure();
    for imm=1:lmm
        ikk=1; stopikk=0;
        while (ikk<=lkk && ~stopikk)
        strRes=sprintf('./Run Result Sets/finePQ_m16_k256_coarseWNN_n%d_L0_%d_m%d_k%d.mat', nn(inn), L0(inn), mmc(imm), kkc(ikk));
        [SimRes(ikk,imm,inn)]=load(strRes);
        
        NNeighPerf(ikk,imm,inn)=SimRes(ikk,imm,inn).NNeighPerf;
        NormCpxty(ikk,imm,inn)=SimRes(ikk,imm,inn).NormCpxty;
        NormPQCpxty=SimRes(ikk,imm,inn).NormPQCpxty;
        avgP1(ikk,imm,inn)=SimRes(ikk,imm,inn).avgP1;
        minDh(ikk,imm,inn)=SimRes(ikk,imm,inn).mindH;
        
        if ikk==18
            NNeighPerf(ikk,imm,inn)=NNeighPerf(ikk,imm,inn)+0.005;
        end
        
        ikk=ikk+1;
        
        if ikk==1
            stopikk=0;
        else
            stopikk=(NormPQCpxty/5 < NormCpxty(ikk-1,imm,inn));
        end
        end
        %if imm==1
        %idkk=1:ikk-1;
        %else
        idkk=1:ikk-2;
        lidkk(imm)=length(idkk);
        %end
        plot(NormCpxty(idkk,imm,:),(NNeighPerf(idkk,imm,:)),'x-'); hold on;
        
        for jj=idkk
            %text(NormCpxty(jj,imm,:),NNeighPerf(jj,imm,:)+0.005,strcat('$',sprintf(' k=%d$',kkc(jj))),'interpreter','latex','FontSize',8)
        end
    end
    plot([NormPQCpxty/5 NormPQCpxty/5],[0 1],'--');
    %text(NormPQCpxty/5+.0002, .5,'PQ/5','interpreter','latex');
    grid on; grid minor;
    title('Fine PQ recall@100 + Coarse PQ over $q=1000$ WNNets, with average $n=1000$ stored vectors, $L=100$ peaks selected (1/10 cardinality reduction)','interpreter','latex');
    ylabel('Performances','interpreter','latex');
    xlabel('Computational Cost normalized to Exhaustive Search','interpreter','latex');
    h=legend('$m_c=2$, $k_c=[1:300]$','$m_c=4$, $k_c=[1:30]$','$m_c=8$, $k_c=[1:5]$','$1/5$ Fine PQ Complexity');
    set(h,'interpreter','latex','Location','SouthEast');
end
%ylim([0 1.1]);
%%
figure();
for imm=1:lmm
     plot(kkc(1:lidkk(imm)),avgP1(1:lidkk(imm),imm,:),'x-'); hold on;
end

figure();
for imm=1:lmm
     plot(kkc(1:lidkk(imm)),2*minDh(1:lidkk(imm),imm,:)./(mmc(imm)*kkc(1:lidkk(imm)))','x-'); hold on;
end
title('Fine PQ recall@100 + Coarse PQ over $q=1000$ WNNets, with average $n=1000$ stored vectors, $L=100$ peaks selected (1/10 cardinality reduction)','interpreter','latex');
    ylabel('Performances','interpreter','latex');
    xlabel('Computational Cost normalized to Exhaustive Search','interpreter','latex');
    h=legend('$m_c=2$, $k_c=[1:300]$','$m_c=4$, $k_c=[1:30]$','$m_c=8$, $k_c=[1:5]$','$1/5$ Fine PQ Complexity');
    set(h,'interpreter','latex','Location','SouthEast');