Author |
Topic: 2008 Postal Tournament (Read 6290 times) |
|
Janzert
Forum Guru
Arimaa player #247
Gender: 
Posts: 1016
|
 |
Re: 2008 Postal Tournament
« Reply #45 on: May 16th, 2008, 6:51pm » |
 Quote  Modify
|
on May 16th, 2008, 2:11pm, Fritzlein wrote:| Janzert, when I compiled the opening statistics, I noticed that Opfor was the most experimental of any player. What is your setup algorithm? |
|
First some motivation,  I wanted to give people variety in OpFor's openings so decided to add more piece setups. I also don't think at OpFor's current level of play that the opening setup matters much, at least as long as it's fairly reasonable.
OpFor randomly chooses between 3 setup styles, Omar, 99of9 or Fritzlein. It then places the major pieces (EMH) and rabbits in the usual positions making sure the elephant is on the same half as the opponent's unless that would be directly across from the oppoent's.1 Then it randomly fills in the dogs and cats in the remaining four slots.
A small bit of self play seemed to show that this at least didn't drastically hurt its play. So I decided to throw it in to mix things up a bit as I was somewhat afraid that people might end up bored by there games with OpFor. I'm glad to see that my fear was completely unfounded and hope that the added setups still gave a little extra spice. 
Janzert
1 As can be seen in the game against sharp, this actually had a bug in it and would not switch when necessary.
|
|
 IP Logged |
|
|
|
mistre
Forum Guru
Gender:
Posts: 553
|
|
Re: 2008 Postal Tournament
« Reply #46 on: May 16th, 2008, 7:10pm » |
Quote Modify
|
on May 16th, 2008, 6:51pm, Janzert wrote:
1 As can be seen in the game against sharp, this actually had a bug in it and would not switch when necessary. |
|
This bug occurred in my game vs Opfor as well. I switched my gold camel/E and it setup its E across from my E. This resulted in a strange opening game, but it only netted a cat for me.
|
|
IP Logged |
|
|
|
Janzert
Forum Guru
Arimaa player #247
Gender:
Posts: 1016
|
|
Re: 2008 Postal Tournament
« Reply #47 on: May 16th, 2008, 8:17pm » |
Quote Modify
|
Oops, meant to reply to this part above as well instead of accidentally cutting it out.
on May 14th, 2008, 4:07pm, Fritzlein wrote:| Janzert, if I didn't know that you hadn't altered OpFor between the Computer Championship and the Postal Mixer, I would insist that you must have souped it up somehow. |
|
Hmm, I hope I haven't given the impression that OpFor in the postal mixer is unchanged from OpFor in the CC. Besides the opening setup work I have made a few small changes (there have been 20 commits to the VCS since the CC version). But by small here I'm definitely measuring by code size/time to write it. At this stage in OpFor I wouldn't be surprised if small changes could still cause large improvements in its play.
The one change that I would not be surprised if it caused a jump in play strength is this. Shortly after the CC ended, it finally sunk into my thick skull what null move pruning actually is and why it helps (i.e. it only took me about a decade or so for realization to strike). I simultaneously realised that what I was doing in OpFor, that I had thought was basically equivalent, was not doing any where near the same thing. This led to me finally adding null move pruning to OpFor and testing showed that it did seem to help although how much I have no idea. This ended up being only 17 lines of code.
In chess I believe null move pruning often adds a couple of hundred elo to an engine. In other games it has not been shown to be useful at all. I have a feeling arimaa will land more on the chess end of the spectrum.
Janzert
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 Postal Tournament
« Reply #48 on: May 17th, 2008, 8:10am » |
Quote Modify
|
on May 16th, 2008, 8:17pm, Janzert wrote:| Hmm, I hope I haven't given the impression that OpFor in the postal mixer is unchanged from OpFor in the CC. |
|
Now that you mention it, you implied in an early post in this thread that you were still developing, and I forgot that bit of exciting news. My comment that I "know" you haven't been making changes was based on your commit log here, which has a last update of March 1. Since I no longer "know" that OpFor is the same as the one that played in the Computer Championship, my grasping for explanations of Opfor's strong performance in the Postal Mixer seems less likely than the obvious explanation that you made it significantly stronger.
Quote:This led to me finally adding null move pruning to OpFor and testing showed that it did seem to help although how much I have no idea. This ended up being only 17 lines of code.
In chess I believe null move pruning often adds a couple of hundred elo to an engine. |
|
Very interesting! I would believe that null-move pruning adds several hundred Elo points of strength to an Arimaa engine, on the theory that the vast majority of moves actually weaken a good position. This is in contrast to Go where very few moves weaken a strong position, and most have some positive value.
Before your explanation I was assuming that OpFor's strong postal performance was a fluke, but now it seems to me there is a genuine possibility that OpFor is 1900 strength postally, and by extension around 2000 strength for 2-minute games and 2100 strength for 30-second games. In my mind the effectiveness of null-move pruning has more ramifications for the future of the Challenge than the number of doublings of thinking time, since I expect that ultimately software breakthroughs will be more important than hardware speedup.
Feng-Hsiung Hsu thinks recursive null-move pruning will be a big deal for Go, but I'm wondering whether Arimaa search even gets to a depth whether the recursion pays off. Is your implementation recursive?
Do we know whether Bomb has null-move pruning?
|
| « Last Edit: May 17th, 2008, 8:21am by Fritzlein » |
IP Logged |
|
|
|
99of9
Forum Guru
Gnobby's creator (player #314)
 
Gender:
Posts: 1413
|
|
Re: 2008 Postal Tournament
« Reply #49 on: May 18th, 2008, 4:16pm » |
Quote Modify
|
on May 17th, 2008, 8:10am, Fritzlein wrote:Do we know whether Bomb has null-move pruning?
|
|
Yes, I'm pretty sure David has previously suggested that it was quite valuable to him.
|
|
IP Logged |
|
|
|
Janzert
Forum Guru
Arimaa player #247
Gender:
Posts: 1016
|
|
Re: 2008 Postal Tournament
« Reply #50 on: May 18th, 2008, 11:15pm » |
Quote Modify
|
on May 17th, 2008, 8:10am, Fritzlein wrote:| Feng-Hsiung Hsu thinks recursive null-move pruning will be a big deal for Go, but I'm wondering whether Arimaa search even gets to a depth whether the recursion pays off. Is your implementation recursive? |
|
No. I don't remember for sure if I experimented with a recursive version or not. At least for current interactive search depths I don't expect that it will make much difference since it's pretty rare to get beyond 3 ply (12 steps) even at postal speeds it seems to be very rare to break 4 ply (16 steps). I did try various adaptive reductions but none of them seemed to do as well as a simple 4 step reduction.
Quote:Do we know whether Bomb has null-move pruning?
|
|
Yes, Fotland has stated that he uses null move pruning as well as a few other pruning strategies (found posts here and here).
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 Postal Tournament
« Reply #51 on: May 19th, 2008, 9:40am » |
Quote Modify
|
on May 18th, 2008, 11:15pm, Janzert wrote:| Yes, Fotland has stated that he uses null move pruning as well as a few other pruning strategies (found posts here and here). |
|
I read each of those posts when written, but apparently it went in one ear and out the other. Thanks for the links. If Bomb was already using null-move pruning and had difficulty coping postally, then OpFor's great start to the Postal Mixer is once again mysterious to me. For now I'm just glad you are still working on it and your work appears to be paying off. I hope that after the postal games your latest version will be available for game room play at least some of the time. Especially I wish you well in the 2009 Computer Championship; your latest additions may already have taken you past Bomb!
|
|
IP Logged |
|
|
|
aaaa
Forum Guru
Arimaa player #958
Posts: 768
|
|
Re: 2008 Postal Tournament
« Reply #52 on: May 20th, 2008, 1:09pm » |
Quote Modify
|
on May 17th, 2008, 8:10am, Fritzlein wrote:| Do we know whether Bomb has null-move pruning? |
|
I had no trouble finding the answer to this question using the search option. You can restrict posts to specific members, but it's not enough just to put a name in the "By User:" field; you also have to select the appropriate option in the pull-down menu right below it.
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 Postal Tournament
« Reply #53 on: Jun 8th, 2008, 4:11pm » |
Quote Modify
|
on May 14th, 2008, 4:07pm, Fritzlein wrote:| I wonder if OpFor isn't benefiting from a bit of complacency on the part of humanity. |
|
Well, there goes the excuse of being taken by surprise. OpFor has beaten both mistre and arimaa_master from unclear positions they thought they were winning, after each had responded to comments about OpFor's unusually good performance. I'm starting to believe that OpFor (including improvements between the Computer Championship and the Postal Mixer) is now clearly the strongest bot ever. I am curious to see, after the Postal Mixer is over, how the improved OpFor does against the old Bomb, and also how it does against humans in fast games.
|
| « Last Edit: Jun 8th, 2008, 4:12pm by Fritzlein » |
IP Logged |
|
|
|
mistre
Forum Guru
Gender:
Posts: 553
|
|
Re: 2008 Postal Tournament
« Reply #54 on: Jun 8th, 2008, 5:22pm » |
Quote Modify
|
What is interesting in both mine and Arimaa_Master's losses to OpFor is that we both had a horse framed. I don't know of any other bot that tries to frame horses and actually succeeds (Other bots attempt but usually don't know how to hold the frame). Later on, OpFor framed one of my dogs and actually made me give it up.
I can't help but wonder if I bungled the end game in my match as I was nearly equivalent materially. OpFor pulled off some surprisingly strong moves down the stretch. Regardless of how I underestimated OpFor, I still thought it would blunder eventually or make some weak moves, but it never did. It's 47b, which I originally thought was a weak move, was actually a very sneaky and strong move that caused me to blunder on 48w and pretty much sealed my chance for a comeback.
|
|
IP Logged |
|
|
|
RonWeasley
Forum Guru
Harry's friend (Arimaa player #441)
Gender:
Posts: 882
|
|
Re: 2008 Postal Tournament
« Reply #55 on: Jun 9th, 2008, 4:48am » |
Quote Modify
|
In my games against OpFor, I felt it left its camel too exposed to my elephant in the opening. I was able to use that to create lasting weakness. I can't say OpFor misdefended the E-H attack because it was trying to defend an offside dog at the same time. The good news is that OpFor set up opportunities for me to blunder that I sometimes didn't see right away. Bomb has that effect on me too, so I'm not sure OpFor is ready to beat Bomb. I would rate OpFor near 1900 and recommend it keep its camel safer in the opening.
|
|
IP Logged |
|
|
|
Arimabuff
Forum Guru
Arimaa player #2764
Gender:
Posts: 589
|
|
Re: 2008 Postal Tournament
« Reply #56 on: Jun 9th, 2008, 6:02am » |
Quote Modify
|
I wonder how well would do the "old" OpFor against the new one. It'd be interesting to pit them against each other or even propose the two versions in the gameroom bot list.
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 Postal Tournament
« Reply #57 on: Jun 15th, 2008, 12:36pm » |
Quote Modify
|
I just noticed the fascinating endgame RonWeasley and mistre are having. On move 72 it is down to EHDCCRR vs. EHDCRR. How much is an extra cat worth in such a position? How should one play when the board is so empty? With so little experience in endgames, I really have no idea, and should not say anything if I did, given that the game is ongoing.
Let me at least say, however, that it could make a huge difference whether draws are still in effect or not. I checked the tournament rules, and they don't say! I would imagine that draws were in effect at the start of the Postal Mixer, although this wasn't specified, and therefore they should be throughout. It would be best to clarify that in case it is affecting the play of either party.
|
| « Last Edit: Jun 15th, 2008, 1:06pm by Fritzlein » |
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 Postal Tournament
« Reply #58 on: Jun 15th, 2008, 12:57pm » |
Quote Modify
|
I count that exactly half of the Postal Mixer games have completed (42 of 84). Meanwhile the two games left that are least progressed are the two between chessandgo and myself; they are on moves 9 and 10 respectively. We did not plan this in order stage a climactic finish, but I hope that's how it turns out!
|
|
IP Logged |
|
|
|
mistre
Forum Guru
Gender:
Posts: 553
|
|
Re: 2008 Postal Tournament
« Reply #59 on: Jun 15th, 2008, 6:54pm » |
Quote Modify
|
on Jun 15th, 2008, 12:36pm, Fritzlein wrote:| I just noticed the fascinating endgame RonWeasley and mistre are having. On move 72 it is down to EHDCCRR vs. EHDCRR. How much is an extra cat worth in such a position? How should one play when the board is so empty? With so little experience in endgames, I really have no idea, and should not say anything if I did, given that the game is ongoing. |
|
Thanks for noticing. I really do feel fortunate that I have got to the position I am in. Ron has played a tremendous game. I have no idea how I am going to play this ending - but it should be interesting. I'll save any actual analysis until the game is over, of course.
|
|
IP Logged |
|
|
|
|
Home
Help
Search
Members
Login
Register
Reply
Notify of replies
Send Topic
Print