.net
All site revenue goes to charity

Subject: towards a detailed exploration of unwinnables

Date: Sun Feb 6 13:54:19 2011
User: Denny
Message:
I've added a link to the Instructions page that points to a code snippet which I think demonstrates how we deal based on game number, variant, and difficulty level. There's a completely open Freecell solver available at fc-solve.berlios.de and I grabbed the code thinking maybe I'd wire it up to our deal generation but then I started thinking about the number of hours of compute time it would likely take to run all our variants and deals through the solver and decided to back off. But if someone else wants to invest the time, we can work out a format whereby I can ingest the results and maybe someday offer a flavor where unwinnables (or at least games not solvable by program) are weeded. It'll be interesting to see what percentage of difficulty 10 6x4 deals are actually winnable. To be useful I'll really need to have results on all the game flavors we currently have stats for I think. Post to this thread if you're going to undertake this and we'll work out a way to communicate. Denny Denny

Link: deal code

Date: Sun Feb 6 14:49:51 2011
User: hotnurse
Message:
Sorry, unrelated but...........I renewed my Premium on Feb. 3rd, cleared my bank of the 4th but still not able to access as Premium member. Thanks, KMc

Date: Sun Feb 6 15:32:17 2011
User: WRAC
Message:
hotnurse, did you re-register? Renewal doesn't happen automatically. See link below. As for unwinnables, my feeling is they should stay. Luck is a part of all card games, albeit to a lesser extent in freecell than many others. Most important, figuring out whether a particular deal is or isn't winnable comprises a large part of the fun and challenge. Freecell, like life, isn't always fair. Learn to live with it. :) In any case, Denny, thanks for posting the code. After all these years, we finally have the details of how the difficulty system works!

Link: Premium Registration

Date: Sun Feb 6 15:39:17 2011
User: TNmountainman
Message:
I still cringe at the 'uncertainty' brought into the statistics of the game of baseball by expanding the season from 154 to 162 games. Based on that, which happened 50(?) years ago - leave it alone....

Date: Sun Feb 6 15:56:16 2011
User: hotnurse
Message:
WRAC, thanks, I didn't complete the process. Sorry!

Date: Sun Feb 6 16:05:18 2011
User: kangaroo
Message:
After the Chat fiasco it's hard to take Denny's proposals seriously any more.

Date: Mon Feb 7 12:02:01 2011
User: RayRay01
Message:
Denny, Thank you for sharing the code! Looks like fun... Regards, Ray

Date: Mon Feb 7 20:08:50 2011
User: ElGuapo
Message:
Would it make the task easier if you used the game stats we already have to narrow down which games in each variant have already been solved?

Date: Mon Feb 7 20:43:11 2011
User: Denny
Message:
Not really. That stats we have would get super sparse on the less common variants and at the higher difficulty levels.

Date: Sun Mar 13 21:56:32 2011
User: free@last
Message:
I've taken a couple of steps. I got your code into a usable format (no small effort). :) Also, I downloaded the solver and tested things out a bit. Not ready to make any promises yet but will update this thread if/when I take another step.

Date: Sun Mar 13 22:24:08 2011
User: Denny
Message:
Cool. I think WRAC also was poking at things a bit. Maybe you two should collaborate. Denny

Date: Sun Mar 13 22:46:52 2011
User: WRAC
Message:
No small effort is right! I worked on it for a bit, but then decided it was more effort than it was worth. Congrats on sticking with it! I don't know if there's anything I can do to help at this point, but let me know if you want to discuss it.

Date: Mon Mar 14 07:58:30 2011
User: free@last
Message:
Thanks WRAC. Next step is to add a command line interface to control the number of columns, freecells, difficulty, etc. AFter that the big question will be how to allow many people to test various sections (probably something simple like a thread here with people saying "I'm testing 8x3-5!" unless someone has a better idea.

Date: Mon Mar 14 09:30:12 2011
User: firenze
Message:
I can't wait for the solver to say a game is unwinnable and have Mastermind solve it. Wait a minute. Mastermind will never see that game if it is eliminated. Actually, he will never again have the pleasure of solving a game that nobody else has solved, because the solver solved it. Sorry, Mastermind.

Date: Mon Mar 14 16:59:01 2011
User: !_--FAST-ISHAM--_
Message:
Nothing really wrong with eliminating unwinnables-just as long as theres one regular set and an "all winnable" set,and the eliminating process doesnt DO anything to the regular games.

Date: Mon Mar 14 19:05:10 2011
User: firenze
Message:
Maybe we should eliminate the "under 10%" games, too. Gosh, I wouldn't want a streak broken by a hard game.

Date: Mon Mar 14 19:43:10 2011
User: Denny
Message:
I'm afraid I'll need to elimnate all games under about 70% to have any chance. If I do anything with this there will for sure be the default mode where you get the full mix. I might have a Premium flavor where you can get winnable-only deals of some sub-set of the variants. Dunno, waiting to see what those guys cook up. Denny

Date: Tue Mar 15 11:28:18 2011
User: free@last
Message:
Still plugging away. Taking me longer to add the CLI because I'm trying checking all the inputs to ensure they are valid. But, to more important topics, I know playing under a previous alias that I played every game type until I had at least one win. I think I played 4x4 over 300 times before I found a winnable one. I can't imagine limiting 4x4 to winnables only. As Denny says, some sub-set might make sense. I can also see some use in tournaments. The worst thing is, when I take a coding break to play a few deals and a particularly tough deal comes up I start thinking "Hey, why not just type it into the solver?" So far I have resisted, even at the expense of a streak. :)

Date: Tue Mar 15 11:34:28 2011
User: firenze
Message:
I don't think there is any doubt that some of the better streaks at this site involved solvers.

Date: Tue Mar 15 13:05:58 2011
User: shiloh
Message:
speaking of unwinnables....they certainly can bring a streak to a screeching halt...add #20843 of bakers dozen to the list that no one can win

Date: Tue Mar 15 18:05:59 2011
User: free@last
Message:
Good news is the I now have a command line configurable generator and a bash shell script (don't ask) that can run a bunch of iterations. The shell interface looks like this: fcs.sh 8x4-11 0 999 which will run the 8x4-11 (duh!) games from 0 to 999. The output looks like this (I used a different variation here to show an example of a "not solved"): Game 8x4 9214-12: This game is solveable. Game 8x4 9215-12: This game is solveable. Game 8x4 9216-12: I could not solve this game. Game 8x4 9217-12: This game is solveable. I expect most people do not have (and do not want) access to a bash shell on their Windows computers. I'll make the script a "normal" program next. The bad news is that I have a decent computer with lots of memory and it took 5 or so minutes to do the 8x4-11 0 999 test. Whether that is typical or not, I don't know.

Date: Tue Mar 15 21:16:27 2011
User: ix
Message:
>> I don't think there is any doubt that some of the better streaks at this site involved solvers. care to elaborate? perhaps you can clear *my* doubts.

Date: Tue Mar 15 21:54:37 2011
User: free@last
Message:
I ran 8x4-11 from 0 to 7999. I got results for 7987 games. Wtf? One of the games was 8x4 2739-11. I played it and it wasn't very hard. (It's actually 3/3 now.) I will investigate further but at this point I don't know what is going on. (Maybe the algorithm gives up after some amount of work?) In any case, I don't think it's a huge deal if some winnable deals don't get counted.

Date: Tue Mar 15 22:39:56 2011
User: EZ-Ed3-LAnseMI
Message:
ix -- I hope firenze isn't talking about my streak of 17 in 10x6.

Date: Wed Mar 16 02:38:30 2011
User: Denny
Message:
That's awesome, free@last. Well done! I'm filled with questions. What's the win rate of some of the tougher variants? Does the solver solve all the 8x4s here? They're proven all solvable by humans. If you leave it grinding what's the approx run time to run thru all 32000 deals of a difficulty level here? I assume that depends a bit on the variant and the difficulty, right? When I did winnable, the winnable games would sometimes pop pretty quick while the solver had to exhaustively keep trying on the hard ones. Can you spare the cycles to run thru all the games here? Can we work out a simple format that I can use to ingest the data? dennis.cronin(a)gmail.com Denny

Date: Wed Mar 16 10:37:21 2011
User: free@last
Message:
For 8x4-11 I estimate 2.5 hours to test all deals. No way I'll be able to do them all myself. Some games it solves almost instantly. Some take 20 or 30 seconds. I haven't tried anything else except one or two deals here and there. I'll send you an email to discuss format.

Date: Wed Mar 16 13:55:13 2011
User: Denny
Message:
So taking 72 variants times the 8 difficulty levels I get 576 groups of 32000 games to process. 10x6 should run really fast as the games are easily solvable. 4x6 will probably require the full 30 secs/game so it may approx balance out But 576 * 2.5 hours is 1140 hours or 60 days so that's approximately 2 months solid runtime on the home CPU. If you do a Windows version I'll run on my home machine. If it's Linux-only I'll run it on the main game server but nice it way down. We'd only need a couple other volunteers to shave it down to about 2 weeks which would certainly be the absolute soonest I'd have the site feature ready. I'm thinking of an option for Premium users to select "winnable" mode and then we'd keep a whole 2nd table of scores. I'd combine game stats as they're relatively big. Drop me an example for 8x4-5 32000 games in format we discussed in email and I'll start working to integrate. Denny

Date: Wed Mar 16 13:58:19 2011
User: Denny
Message:
In email free@last points out the 8x4-5 will likely not be a good test since all those are known to be winnable. I've asked for 13x0-5 instead.

Date: Wed Mar 16 13:58:59 2011
User: Denny
Message:
P.S. free@last was generously quite polite and didn't outright call me "stupid". :-)

Date: Wed Mar 16 14:33:42 2011
User: firenze
Message:
If 8x4s are all solvable, couldn't you assume that 8x5, 8x6, etc. are also solvable and not have to do them? The same with other like 9x4s, 5s, 6s, and 7s. Wouldn't that save some computer time?

Date: Wed Mar 16 19:12:11 2011
User: free@last
Message:
Let me catch the interested parties up. I just sent Denny an example file with the 10x1-5 list of non-solved. I found out when I tried to run 13x0 that the binary for the solver is hard coded to only allow up to 10 columns. I'll need to download the source and recompile to do the wider games. firenze, that is correct. Since Denny wants a list of unsolvables I think the easier thing will be for me to start with x0's and then feed that list in to x1 and that result into x2 etc. Just for example, 10x1-5 had <600 unsolved which should reduce the time for x2 enormously. Thanks for pointing this out!

Date: Thu Mar 17 03:35:13 2011
User: ElGuapo
Message:
Denny wrote: "I'm thinking of an option for Premium users to select 'winnable' mode and then we'd keep a whole 2nd table of scores. I'd combine game stats as they're relatively big." If you start a whole 2nd table *from scratch*, then people are always going to want to toggle back to the original list for comparison. That's too unwieldy. I think what Premium players would actually want to see is these new winnable-only streaks incorporated into the existing scores pages. To do this you could add a dynamic link that toggles back and forth between the two modes without requiring the page to reload. See link below for an example on the ratings site where we do the same kind of thing. The only trick here is you'd have to code it such that all players now have two IDs potentially: their Random Deal stats, and their All-Winnable stats. Maybe you color code the All-Winnable persona or something so it stands out from the rest.

Link: See the "Rank All" link at top right of this page

Date: Thu Mar 17 10:39:54 2011
User: firenze
Message:
I would guess that there won't be a lot of variants in the Premium winnables format. Probably less than half of what we have now. Most of the variants that are 13 sum or more and seven columns or more may be all-winnable now with the exception of 13x0 and a couple of games in 12x1. Also the 10 sum or less may not produce enough winnables to justify entry in the Premium winnables format. I would estimate the number of variants in the new format will be under 30 with the 11-12 sums and 13x0 being the most popular.

Date: Thu Mar 17 12:35:06 2011
User: free@last
Message:
Another update: After nearly 4 painful hours, I managed to get the solver recompiled allowing up to 13 columns and 10 freecells. I ran all the 13x0 difficulties over night and then promptly deleted all the results. D'oh! They are re-running now. I will use the unsolved list from these to run run 13x1's next, as firenze suggested. Assuming I don't continue to delete all my work, I think I run most of what might be interesting in a couple weeks. No pressure, Denny. :)

Date: Thu Mar 17 13:06:52 2011
User: Denny
Message:
firenze makes good points. I was wondering what the experience should be where the games are all winnable anyhow. If you play an 8x4 and select "winnable" mode, does the server need to know that in that case it's all the same stats? Denny

Date: Thu Mar 17 16:41:44 2011
User: ElGuapo
Message:
Right, I think you'll need to give the server a list of variants where unwinnables exist and only create separate scores for those.

Date: Thu Mar 17 18:00:02 2011
User: free@last
Message:
Instead of eliminating the unwinnables, leave them as is and if/when someone plays one and loses, don't count that game against their unwinnable streak but end their normal streak? Also, do you plan to remove games from the unwinnable list if someone wins one? Just wondering.

Date: Thu Mar 17 22:13:03 2011
User: free@last
Message:
I've got some interesting results so far. I've finished 13xn. Here's how many games could not be solved for difficulty 5-12 of 13x0: 521 690 720 888 1137 1408 1654 1789 1846 And here are the results for 13x1: 0 0 0 0 0 0 0 1 All 13x2's (and up) are solvable by the program.

Date: Thu Mar 17 22:58:56 2011
User: TNmountainman
Message:
Not to butt in, but you've got a list of 9 numbers (totals per 'n') for 13x0, and yet "5-12" should only result in 8 numbers. For 13x1, you've got 8, as it should be.

Date: Fri Mar 18 03:04:59 2011
User: ElGuapo
Message:
And now that you mention it, only difficulty levels 5–10 need be looked at for this project since it's only relevant to streaking. Very interesting to see actual numbers for the unwinnables, free@last.

Date: Fri Mar 18 07:35:45 2011
User: free@last
Message:
Ah, yes. Should have been: (cut & paste error) 521 690 888 1137 1408 1654 1789 1846 Actually, I think you can manually ask for -11 and -12 randoms and have them count for your streak. More important would be for my computer to not reboot in the middle of the night to install urgent security updates such as the disabling of Autorun! Grrr. No further significant progress last night.

Post follow-up
Username: New user? Create a free account here
Password: Note: username and password are case-sensitive
Message:
Editor by summernote.org
Email notification:

All content copyright ©2024 Freecell.net
By using our games you consent to our minimal use of cookies to maintain basic state.
Maintained by Dennis Cronin