Good. Which is really the question in all of this. Finding good heuristics. And the
solution I'm, and so that we had two things we talked about. One is actually
different heuristics actually matter as we can see via these numbers. And we can
find good heuristics by cheating. Just relaxing the problem, which is just what
cheating is. And so if we solve a relaxed problem instead of the original
problem, we get to the solutions faster. And the result of those searches might
actually be a good heuristic depending on how much we cheat. So the wonderful
thing here is if we relax the problem and use the real solution to the relaxed
problem as a heuristic to the original problem, then we're underestimating
automatically. This gives us admissible heuristics automatically. Why? Because any
solution to the real problem is also a solution to the relaxed problem. So we're
underestimating. So this is a good way of doing it. And if you don't know how to
cheat, learn how to cheat. Okay? It gives you good heuristics. And sometimes the
relaxed problem is still too difficult to solve. Then you need to cheat more. Okay.
Who of you looked at this? You should. It gives you very good intuitions about
search. And it's much better than counting sheep in certain
situations. I think you should really look at it. Just looking at this
little snake of where you're all going and so on and what you've searched and
how often you search these things, it actually tells you a lot about what
search is. So it has a depth-first search, a breadth-first search, an A star search
for various heuristics setting. And you can just play with them a little bit. You
don't have to look, you shouldn't actually for certain things wait until
it found, it actually found a solution because that takes too long. In some
cases too long for the server that this runs on to exist or you to exist or
whatever. But look at that. Really do it. It gives you an intuition. By the way,
we're probably going to start game search or politely expressed
adversarial search, right? Solving problems by game, solving games where you
have an adversary by search and building chess systems or so. Where these
kind of things matter because it's just a kind of search on steroids where you
have an uncontrolled environment. But many of the intuitions still apply.
you
Presenters
Zugänglich über
Offener Zugang
Dauer
00:04:22 Min
Aufnahmedatum
2020-10-28
Hochgeladen am
2020-10-28 09:57:03
Sprache
en-US
Recap: Finding Good Heuristics
Main video on the topic in chapter 7 clip 9.