User:DroneBetter/esoteric search program knowledge

this page is intended to store information and quotation for referenciation and also a small amount of opinionation in humoracious juxtapositation

all quotes herein are true and real and unadulterated and I may be trusted

observations

• qfind's width-exhausting capability is surpassed by JLS's at high speeds and low denominators, both seem to be surpassed by ikpx2 for extremely low denominators (ie. c/2 in rules such as 2×2 where they are hard to find), and LSSS (nonexhaustive though it may be) is good for finding examples of speeds when one doesn't care about width but existence

Darren Li

• despite the fact that it says it is 'adaptive widening,' ikpx2 moving on from a width does in fact imply it has been searched exhaustively (and, due to its concatenation of subsections of different symmetry, in fact means it has searched a superset of the width)
• as far as SAT-solver-utilisers go, ikpx2 is better than LLS due to 'automation of the usage of said SAT solvers'

“ ikpx2 is literally just qfind but with ridiculously high deepening, and qfind is just gfind with ridiculously high deepening ”

“ the pecking order for spaceships is definitely ikpx2 > qfind > jls ”

“ Once [qfind's] deepening exceeds like, 50, ikpx2 is unilaterally faster - the only problem is that ikpx2's memory usage is unbounded ”

“ qfind is [expletive] cracked actually i need to study how its lookahead works under a microscope ”

wwei

• generally, JLS is better than ikpx2 for odd-period diagonal spaceships since it is capable of exploiting diagonal symmetry (not implemented in ikpx2), but ikpx2 is better for even ones since it is capable of diagonal glide-symmetry (apparently, JLS is also, but it only works for oscillators)

“ JLS and ikpx2 understand how to propagate the consequences of a guess to a row that qfind hasn't even considered yet ”

apg (per wwei)

• the algorithm JLS uses is DPLL

Sokwe

• how to avoid small objects congesting searches when looking for nonminimal results:

“ Specifically, when [Nicolay Beluchenko's version of WLS (WLS 6.3)] finds a complete spaceship, it will only continue on that branch if the subsequent pattern interacts with the spaceship. For example, it can be used to find complete sets of spaceships for certain combinations of speed and width/height (when that set is finite). ”