In XQDA.m ,W is [d,r] , why r is chose from latent>1？

[Deeplearning20]

159 r = sum(latent > 1);
160 energy = sum(latent(1:r)) / energy;
In XQDA.m ,I know W is [d,r] , but why r is chose from latent>1？
Why is the original dimension d=26960 unchanged? Has the dimension been reduced after XQDA?
How do I adjust r if I want to improve the performance of metric learning?
thanks very much!