|
||||
|
Title: Database of unique wins in 2 Post by aaaa on Oct 8th, 2012, 7:52pm Watch this category (http://arimaa.com/arimaa/mwiki/index.php/Category:Unique_wins_in_2) as I fill the wiki with positions in puzzle format in which there is an (almost surely) unique move that forces goal or elimination in two (it's theoretically possible a solution isn't unique because of the possibility of immobilization). Feel free to edit these articles, e.g. to embellish specific puzzles in some way. |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 9th, 2012, 10:12am on 10/08/12 at 19:52:49, aaaa wrote:
Wow, good job. The uniqueness is itself good filtering condition. I have opened 2 of them and they are really easy. Feature to mark easily the complicated cases would made it even better. (Of course I often miss moves which look easy for browni, someone could consider most of them difficult). I will definitely look at it more closely. Unfortunately the positions are not shown on IPhone:(. At least on my one. Link to the static diagram would help. Wow 16:16s is nice, was your solution generated by bot? It seemed to me blocking f5 was crucial and hf6 re8 defends otherwise. Oh I have probably wrongly read the solution ... it definitely provides several false solutions ... good composition :). (I have used bot to check the solution is OK and what are the answers to the false variants.) I am going to add it to the puzzle collection ;) |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 9th, 2012, 12:24pm on 10/09/12 at 10:12:15, Hippo wrote:
Leo's dynamic image creation tool seems ideal to remedy this. Anyone have any particular preferences for board size and file format? Otherwise, I'm inclined to go for the maximum (500) and JPEG, respectively. Quote:
I think it's a bad idea to do this repeatedly. Although positions derived from games have been added manually in the past, I expect something like a five-digit number of unique wins-in-2's and it would be nice to keep the existing puzzle collection separate so that it won't be too hard to find differently created puzzles in the future. |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 9th, 2012, 12:59pm on 10/09/12 at 12:24:07, aaaa wrote:
OK seems the unique solution makes most of them interesting and there is no need for filtering. I went to game 134 so far and I still love 16 most thanks to the interesting defense of the simillar variants. I have a feeling of a composition.... 154/31 is nice as well ;) |
||||
|
Title: Re: Database of unique wins in 2 Post by Janzert on Oct 9th, 2012, 2:42pm on 10/09/12 at 12:24:07, aaaa wrote:
The wiki diagram template (http://arimaa.com/arimaa/mwiki/index.php/Template:Arimaa_diagram) (as used in past event reports (http://arimaa.com/arimaa/mwiki/index.php/2012_World_Championship,_Round_1)) could also be an option. Janzert |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 9th, 2012, 3:17pm I can't imagine anything but links because of how already large the articles are. |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 10th, 2012, 7:02am One more thing ... the solutions don't show in the flash view, I can read them only howering on the link. |
||||
|
Title: Re: Database of unique wins in 2 Post by Janzert on Oct 10th, 2012, 2:47pm on 10/10/12 at 07:02:37, Hippo wrote:
I believe you have to go to the solution link, then press the "solution" button on the left hand side (next to the viewer, not part of it) to actually get the solution movelist. Then press to "Show" in the game viewer to display it on the board. |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 10th, 2012, 6:16pm on 10/10/12 at 14:47:39, Janzert wrote:
Oh thanks. In that case I don't understand why there are 2 links on the page ... why not using only the solution link. Before the solution button is pressed the solution remains hidden ... and solution button on the first link does nothing. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 10th, 2012, 9:35pm The reason why I did it this way was because the link that allows one to see the solution already gives it away in its URL. |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 11th, 2012, 3:50am on 10/10/12 at 21:35:16, aaaa wrote:
Wow, I am blind ... . Get/Post problem in the cgi you don't have under controll. One walk around would be use url's of the form adding long enough dummy constant to the url (http://arimaa.com/arimaa/puzzles/auto/showPuz.cgi?g=304&m=23b&w=2&dummy=this%20is%20a%20very%20long%20constant%20to%20hide%20the%20solution%20from%20the%20visible%20part%20of%20the%20url.%20As%20we%20are%20afraid%20of%20usage%20of%20very%20wide%20screens,%20we%20must%20use%20this%20constant%20really,%20really%20long.%20May%20be%20this%20is%20already%20sufficient.&s=rb6w+cc6x+ra6s+ra4s+ra3s) 368 has nice motif. ... Now I like using your solution links. I just don't watch the url till I am not satisfied with my solution ... than I just look at the link not replaying the solution. So this link form has it's value. ... What about using the dummy constant trick in left link and preserve the solution link? |
||||
|
Title: Re: Database of unique wins in 2 Post by Fritzlein on Oct 11th, 2012, 8:02am on 10/08/12 at 19:52:49, aaaa wrote:
Thanks aaaa. This is a significant step forward from Omar's first pass at automated mining of the game database for goal puzzles. on 10/11/12 at 03:50:08, Hippo wrote:
If some people are willing to work through all the problems and point out the interesting ones, that is a service to those of us who enjoy a few good puzzles but don't have the commitment to do them all. I can see value in an overlay page or subset page filtered by human judgment. |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 11th, 2012, 12:11pm on 10/11/12 at 08:02:29, Fritzlein wrote:
Actually after first randomly opened easy puzzles, I am going through all the puzzles (now at 400). Only rarely the puzzle is really easy. The puzzles when all 4 steps are made by one rabbit (or when two rabbits on empty board reach 4th rank) seemed to be generaly easy, but I must say in one puzzle I thrown away such solution at the first glance and returned to it much later. Sometimes the puzzles from the same game have the same solution or at least solution with the same motif. The unique solution condition made very good prefiltering. Ohh, aaaa declared hardware problems. I just hope it was not due to owerheating while preparing all these puzzles. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 11th, 2012, 1:49pm on 10/11/12 at 03:50:08, Hippo wrote:
Not only is this a really ugly hack, the end of the URL will still show up in the status bar of certain browsers when hovering over the link. |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 11th, 2012, 4:05pm on 10/11/12 at 13:49:54, aaaa wrote:
Ugly ugly, question is what you want to achieve ;). Problems with the tail could be solved by the same hack ... just add &dummytail= ... but if a browser shows the whole link split on several lines, you can only hope the move is sufficiently hidden in the garbage ;) ... you can as well hide s= among other z= x= with a arimaa move like context ... I agree that post would be much better for these purposes ... 562:49g is goal in position I would not expect to happen. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 12th, 2012, 10:01am I've manually changed one puzzle page (http://arimaa.com/arimaa/mwiki/index.php/Unique_wins_in_2/2002) to how I currently think the puzzles should be presented. Comments? |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 12th, 2012, 12:35pm on 10/12/12 at 10:01:37, aaaa wrote:
Images work well on Iphone, I like it. I would mainly use puzzle image link on it. On comp I would use the solution link to flash ignoring URL till I decide to read it ;) as I do so far. I have looked at puzzless till game 600. Often the same motif repeats (as one could expect in data mined puzzles). Next step could be to detect which part of the board was "defining" the puzzle. ... and select unique puzzles upto it. But how to formalize this? (After detecting important part of the board ... only relative strength of pieces which could influence each other on the part of the board is important for the classification, mirror/central symmetry is not important as well.) I don't ask you to go for this step, and I am not adding it to my TODO list either. May be it could be a way to make big step in static goal in 2 detection ... but I am afraid the set of patterns would be too huge for it. (The most often repeated patterns would be more interesting than patterns appearing only rarely ... as single rabbit runs usually depend on huge part of the board it would more probably be unique. More important patterns would be on smaller part of the board and therefore more often to be seen. On the other side puzzle like 16:16s would be probably eliminated by prunning rare patterns.) On the other side it would be interesting to have the positions/solutions classified by the pattern. Hmmm ... classifiyng by solutions rather than puzzles would be better. I am not sure the single rabbit runs would not be repeated often but "joining them together" would be as good as filtering them out. ... 766 I like the attack/defense balance. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 23rd, 2012, 4:03pm OK, there are now almost twenty thousand puzzles in that format on the wiki. Enjoy. |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 23rd, 2012, 4:11pm on 10/23/12 at 16:03:50, aaaa wrote:
Yes this is for years of study ... . One question ... I thought ... http://arimaa.com/arimaa/puzzles/auto/showPuz.cgi?g=172760&m=35b&w=2 should belong to the collection as well. Have I missed something? |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 23rd, 2012, 4:26pm You didn't. Only a few positions were checked per game as the data mining would have taken an exorbitant amount of time otherwise. |
||||
|
Title: Re: Database of unique wins in 2 Post by Hippo on Oct 23rd, 2012, 4:39pm Oh, I expected this is full set of unique wins in 2 from the game database. OK, good to know. How did you filtered the positions ... your bot did random sampling or your bot guessed for each position if there would not be goal in 2? The latter case would be serious problem as the most surprising positions would be missed in that case. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 24th, 2012, 9:58am I didn't do anything very sophisticated; I checked Omar's earlier generated list of positions and the position before the last non-finishing move by the winner in every game. I also got 4 more puzzles after treating each side as the winner in abandoned games. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 25th, 2012, 1:36pm Regrettably, I just discovered that zugzwang wins weren't caught by my approach and these must surely be considered interesting puzzles. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Oct 30th, 2012, 8:13pm I've updated the database. Handling zugzwang correctly not only resulted in more unique wins being found, it also exposed some existing solutions as not being unique after all. In addition, I've added every conforming puzzle on the original puzzle page that wasn't already discovered automatically. |
||||
|
Title: Re: Database of unique wins in 2 Post by omar on Mar 2nd, 2013, 12:06am Wow, thanks for creating this aaaa. It's pretty fun going through these puzzles. Martti and I are working on a puzzle app for the iPhone. They are not typical "win in x" type puzzles. Instead you are just put in positions where you should be able to win. The quicker you win the more points you get. Some win sequences can take even up to 10 moves. We are not concerned about optimal move sequences, nor does the bot play an optimal defense. It's more like bot bashing from given starting positions. The app is not for hard-core Arimaa players and is more for people who like to play casual puzzle games like 'Cut the rope'. Hopefully it will introduce more people to the Arimaa rules. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on May 4th, 2013, 9:02am I've stumbled on a generated puzzle that is a false positive on account of the immobilization victory condition: On move 42g of game 262074 (http://arimaa.com/arimaa/gameroom/comments.cgi?gid=262074), there is only one move that forces goal or elimination in 2. However, if Gold makes any move that doesn't change the position in the northern half of the board, zugzwang would force Silver to allow immobilization on the next move. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on May 8th, 2013, 3:49pm By performing an immobilization check on the antepenultimate half-move of each game that actually ended that way, 95 puzzles were exposed as not having a unique solution. If anyone knows of other specific puzzles they suspect are actually non-unique, please let me know of them. |
||||
|
Title: Re: Database of unique wins in 2 Post by Fritzlein on Aug 3rd, 2014, 3:57pm Thanks again for doing this, aaaa. I have just now started using the database for real, after yet another of my typical fail-to-find-the-short-win games. It's an invaluable tool. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Dec 2nd, 2014, 2:22pm Going through the games in their entirety appears to be a tractable proposition after all. It will probably take me in the order of weeks to process the existing database. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Dec 4th, 2014, 6:30pm on 05/08/13 at 15:49:00, aaaa wrote:
Using this condition to also look for false negatives, i.e. positions that are unique wins-in-2 exactly because of the possibility of immobilization, doesn't seem to take up too much of the total search time. That means the database will no longer be restricted to puzzles in which it is goal and/or elimination that can be forced. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Jan 18th, 2015, 10:10pm I've finally caught up with the current state of the database (not counting the current incomplete month). The count now stands at 99959 puzzles and that's not mentioning the fact that they underwent a more scrutinizing process than before and that more false positives have been removed as well. |
||||
|
Title: Re: Database of unique wins in 2 Post by Janzert on Jan 19th, 2015, 11:19am Very nice, aaaa. Janzert |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Jan 25th, 2015, 2:43pm I've extended the strict test to the third-to-last move from any immobilizing player. This added 109 puzzles and removed 2, bringing the total into the six digits. |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Mar 11th, 2015, 12:13pm By writing a function that, in the vast majority of cases, can rule out the possibility of immobilizing the opponent with the current move, I've been able to mine the games for all unique wins-in-2 (except those involving repetition) in a reasonable amount of time. With possibly a few exceptions, the puzzle database should now be both complete and correct, including with respect to immobilization. In addition, I've been maintaining a scoreboard of some sort (http://arimaa.com/arimaa/mwiki/index.php/Unique_wins_in_2/Scoreboard), which, for each player, shows how often they actually played the correct move when these positions occurred and, probably more informatively, the most recent such puzzle they didn't solve over the board. |
||||
|
Title: Re: Database of unique wins in 2 Post by Manuel on Mar 12th, 2015, 4:56am That scoreboard is nice (although I don't like my position on it...). Could you also add a column with number of games lost after missing the unique win-in-2? (and a button to exclude bots from the table would also be great btw). |
||||
|
Title: Re: Database of unique wins in 2 Post by aaaa on Jul 21st, 2019, 5:38pm There is now also a dedicated list of zugzwangs (http://arimaa.com/arimaa/mwiki/index.php/Unique_wins_in_2/Zugzwang). |
||||
|
Arimaa Forum » Powered by YaBB 1 Gold - SP 1.3.1! YaBB © 2000-2003. All Rights Reserved. |