Thanks very much for interesting in our work and sorry for this empty github repository. The fundamental algorithm is quite easy to implement: just stop the linear probing earlier far from converge. The code used for our experiments will be released soon (I guess it should be ready before the end of April, sorry for the procrastination ... Ahha, two years later now, really don't want to release those ugly code T_T. Anyway, the theory and idea of this paper is indeed simple enough for anyone to write their own code, as suggested by one of the reviewer, lol.) But the algorithm is really simple, so why not look deeper into the proofs for the toy model :-)