Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Planner programming blows my mind

hillelwayne.com

Post reply on HN
Print view View original

1–10 of 75 posts

Re: Planner programming blows my mind

#3

Post by sterlind »

I've actually used Picat's planning mode at work!

I prototyped a system to orchestrate maintenance on fleets of devices. The idea was that, rather than telling the system how to do it (e.g. workflows to roll out an update), you'd tell the system what you wanted (e.g. up to date machines), what actions were available (e.g. pull a machine from rotation, apply an update), and what constraints to obey (e.g. X of Y machines must be online, don't work in more than two regions simultaneously.)

I modeled a few scenarios like that in Picat and had it generate optimal plans. It worked swimmingly for pet problems, but predictably fell over scaling to cattle sizes. Planning is EXPTIME after all (e.g. Towers of Hanoi).*

Picat does have an escape hatch - you can define heuristics - so I built a random forest of state predicates and trained a naive Bayes classifier to predict fruitful paths. But even with that, and symmetry breaking constraints, and even some hierarchical planning, I couldn't make it work without too much handholding.

It's still AI winter for the classic GOFAI problem domains, apparently. :/

* maybe not, actually, if you reformulate the planning problem as returning a polynomial-time generator of a potentially exponentially long plan

Re: Planner programming blows my mind

#5

Post by polivier »

sterlind wrote: post #3

I've actually used Picat's planning mode at work! I prototyped a system to orchestrate maintenance on fleets of devices. The idea was that, rather than telling the system how to do it (e.g. workflows to roll out an update), you'd tell the system what you wanted (e.g. up to date machines), what actions were available (e.g. pull a machine from rotation, apply an update), and what constraints to obey (e.g. X of Y machin…

It is possible that something like CP-SAT (https://developers.google.com/optimization/cp) would have scaled well in your case. This solver easily handles an absurd amount of variables and constraints, and has excellent heuristics built-in.

Re: Planner programming blows my mind

#8

Post by nickpsecurity »

sterlind wrote: post #3

I've actually used Picat's planning mode at work! I prototyped a system to orchestrate maintenance on fleets of devices. The idea was that, rather than telling the system how to do it (e.g. workflows to roll out an update), you'd tell the system what you wanted (e.g. up to date machines), what actions were available (e.g. pull a machine from rotation, apply an update), and what constraints to obey (e.g. X of Y machin…

"Planning is EXPTIME after all (e.g. Towers of Hanoi).*"

The old planners would include meta-rules, or heuristics, that decided which rules to apply. That would cut the search space. Some split the problem into different representations with specialized, automated solvers. Jahob Analysis System and Cyc come to mind.

Far as for real-world use, the neatest design I remember from classic A.I. was the Procedural Reasoning System. I've always wanted to see a version of it rebuilt with modern methods supplementing its weaknesses. Just for kicks and to see what it could do.

https://en.wikipedia.org/wiki/Procedural_reasoning_system

Re: Planner programming blows my mind

#9

Post by OJFord »

sterlind wrote: post #3

I've actually used Picat's planning mode at work! I prototyped a system to orchestrate maintenance on fleets of devices. The idea was that, rather than telling the system how to do it (e.g. workflows to roll out an update), you'd tell the system what you wanted (e.g. up to date machines), what actions were available (e.g. pull a machine from rotation, apply an update), and what constraints to obey (e.g. X of Y machin…

So it's not actually different (in result/performance) from the depth-first search you'd get naïvely mashing some keys dimly recalling prolog?

Re: Planner programming blows my mind

#10

Post by 7thaccount »

sterlind wrote: post #3

I've actually used Picat's planning mode at work! I prototyped a system to orchestrate maintenance on fleets of devices. The idea was that, rather than telling the system how to do it (e.g. workflows to roll out an update), you'd tell the system what you wanted (e.g. up to date machines), what actions were available (e.g. pull a machine from rotation, apply an update), and what constraints to obey (e.g. X of Y machin…

There are plenty of commercial solvers out there that beat the pants off the open source options in terms of performance and in terms of depleting one's wallet :)

CPLEX, Xpress, GUROBI, and Hexaly all come to mind. Hexaly is really good for scheduling problems and things like vehicle routing. You typically access these via an API they offer you for the popular industry languages. This approach seems to make a lot more sense to me than having a dedicated solver language that isn't as good for all the general purpose stuff. Calling GUROBI from Python is a breeze as is all the standard stuff in Python. Mosek is a lot cheaper than GUROBI, but both of it's APIs are extremely low level and the performance isn't as good as GUROBI either.

Post reply on HN