Jump to content


Photo

Probe Stratego a Good Single Player game


  • Please log in to reply
68 replies to this topic

#41 Moriarty

Moriarty

    Major

  • Members
  • PipPipPipPipPipPipPipPipPip
  • 1,178 posts
  • Coat of arms
  • Gold Colonel

Posted 23 January 2016 - 01:39 AM

whenever i try downloading it it says it is a file that might harm my computer...

and google chrome blocks it


I have no idea what that thing under my avatar is. I've always liked stickers

#42 sevenseas

sevenseas

    Major

  • Banned
  • PipPipPipPipPipPipPipPipPip
  • 1,037 posts
  • Coat of arms
  • Platinum Marshal

Posted 23 January 2016 - 02:01 AM

whenever i try downloading it it says it is a file that might harm my computer...
and google chrome blocks it


Some malicious executable files can harm your computer but pretty much all are safe, especially if you trust the source.

It is perfectly safe to download :)
I play as Sevenseas & Don't Cry

#43 Moriarty

Moriarty

    Major

  • Members
  • PipPipPipPipPipPipPipPipPip
  • 1,178 posts
  • Coat of arms
  • Gold Colonel

Posted 23 January 2016 - 02:09 AM

yes, but chrome prevents me from doing so


I have no idea what that thing under my avatar is. I've always liked stickers

#44 tomato123456

tomato123456

    Bomb

  • Members
  • PipPipPip
  • 51 posts
  • Coat of arms
  • Silver Miner

Posted 23 January 2016 - 09:32 AM

Is it true that Probe will remember your setups and use that information against you if you play the same setup again?

 

No, Probe does not have that ability. There are 50,000 setups that Probe can choose from and the program picks one randomly unless you specifically choose one setup to play against. That last option may be handy if you want to test a certain strategy or setup yourself, but if you just want to play against the program you will be randomly matched with one of the 50,000 options out there.



#45 Imer

Imer

    New Recruit

  • Members
  • Pip
  • 9 posts
  • Coat of arms
  • Bronze Spy

Posted 24 January 2016 - 01:21 AM

Hi, I wrote Probe and would be happy to answer any questions about it.

 

For starters, the folks at Jumbo contacted me several years ago about using Probe as their AI engine, and then never followed through. So blame them if their engine is crap, which is disappointing to read here. I would have been happy to improve Probe and license it to them.

 

Vincent de Boer wrote his original AI, Invincible, as a student project in a short period of time. It played remarkably well for as little effort as he had put into it, but wasn't nearly as good as Probe or Master of the Flag. It didn't use recursion, which is probably fatal for an algorithm to play a turn-based game. Also, I personally don't believe a computer program should try to pursue a plan-oriented approach to Stratego just because humans do. All this said, Vincent is a very bright guy, and obviously a superb player himself, so if he has continued to work on his AI in the years since Invincible, it's probably quite good today.

 

Master of the Flag is an excellent program. in the computer world championships Probe was able to beat it (except for one year when Probe had a bug), but only just barely. Master of the Flag plays conservatively compared to Probe, and I always wondered which would be the more interesting opponent for a human over time. I don't know if its author has continued to work on it.

 

There used to be a wiki that discussed these programs and detailed the tournament results, but Jumbo made its supporter take it down because it used the name StrategoUSA. Too bad.

 

Probe is indeed capable of remembering whatever it learns about your setups for future games and using that information against you through pattern matching, but that feature is off by default on the downloadable version. I put it in because too many players on Metaforge were lottoing against the Probe engine there and then quitting if they fell behind. This was my revenge.

 

Speaking of learning, I am convinced that a Stratego AI capable of defeating top human opponents must be a learning program that amasses statistics about players, both individually and collectively. I even published a journal paper to this effect some years back. I say this because I don't believe a computer program has any particular advantage over a human besides the ability to detect and exploit bias. Stratego isn't chess, where a program can just run through millions of combinations with perfect certainty of the result. Along these lines I did analyze thousands of setups used by players on Gravon and built tables for the likelihood of the use of certain types of setups (which among other things revealed an amusingly strong right hand bias). Probe uses these tables, but they're static. Probe also knows how frequently players position individual pieces on specific squares of the board for setups, both for absolute and relative positions, and builds a probability table from this data. But why not a program that also, say, learns how frequently its opponent moves its Spy for the first time within x moves after the program's Marshal is revealed? There are many useful patterns like this to detect and a computer program should be able to perform this task much better than a human. Most of the complaints about Stratego AIs is that they are predictable, rather than just plain stupid (this site's AI notwithstanding, apparently). Well, the way to be effectively less predictable is to learn, to adapt. Stratego AI's need to be able to do that.

 

Hmm what else? Probe does not bluff. Strictly speaking, bluffing doesn't exist. Instead, there is the probable outcome of a move by a piece whose rank the opponent doesn't know. That should be revealed during recursion by the likelihood the opponent will not risk a strong known piece against an unknown piece that might be able to capture it. In other words, it may be safe to make a suicide move if your opponent doesn't realize it's a suicide move, and you might want to make that move to push that piece away. Probe does that inherently. On the defensive side Probe recognizes moves that suggest an opponent piece's rank, but not every time and not if subsequent moves undermine that assumption. All this is a dance of probabilities, and to beat the point to death, a successful Stratego AI needs to be able to modify these probabilities over time. There are humans who have played literally thousands of games against Probe, during which time the human has learned a lot about Probe, while Probe has learned absolutely nothing about him.

 

Someone mentioned Probe not defending an open Flag. The problem with guarding a Flag too zealously, of course, is that you telegraph its location. It's a balancing act. Tip the scales too far in the other direction and the complaint would have been that Probe doesn't know how to make an open Flag escape notice. So, too, Probe becomes more aggressive if it believes its Flag to be at risk, but it can't do that too obviously either, or it will become apparent that something is amiss. Another balancing act. Whoever in this thread suggested that Probe lottos must have been on the verge of capturing Probe's Flag because Probe never plays kamikaze Stratego for the heck of it the way bored humans sometimes do.

 

Thanks for reading. I'll check in again on this forum from time to time. Or you can e-mail me directly, if you wish, by clicking Support:

 

http://www.probe.imersatz.com/

 

 


  • GaryLShelton, The Prof and Moriarty like this

#46 Nortrom

Nortrom

    Captain

  • WC Online Team
  • 823 posts
  • Coat of arms
  • Platinum Marshal

Posted 24 January 2016 - 01:23 AM

Imer,

 

Nice to see you're still around. Are you still updating Probe? ( You might know me as lightwing instead ;) )


"Rock is overpowered, paper is fine" - scissors

 

Follow the 3rd Online World Championship: http://forum.strateg...d-championship/


#47 Imer

Imer

    New Recruit

  • Members
  • Pip
  • 9 posts
  • Coat of arms
  • Bronze Spy

Posted 24 January 2016 - 05:42 AM

Lightwing! Congrats on yet another WC! 4X, quite a record, I am in awe.

 

I have not worked on Probe in a long time. It reached a point where I decided a complete rewrite was required to achieve any significant gains, and I wasn't up for that, especially with all the changes in the Stratego-branded Internet world. I do wish Probe ran on mobile devices, but that too would take an effort. We'll see. I've never played the AI on this site, but it's discouraging to read the reports on it. The world deserves better.

 

Good to hear from you. Stay well.



#48 chickenofdoom

chickenofdoom

    Spy

  • Members
  • PipPip
  • 10 posts
  • Coat of arms
  • Bronze General

Posted 24 January 2016 - 08:22 AM

Where is the option for Probe to learn your setups? I haven't been able to find it.



#49 GaryLShelton

GaryLShelton

    Marshal

  • Moderators
  • 4,265 posts
  • Coat of arms
  • Gold Captain

Posted 24 January 2016 - 03:09 PM

Master of the Flag is an excellent program. in the computer world championships Probe was able to beat it (except for one year when Probe had a bug), but only just barely. Master of the Flag plays conservatively compared to Probe, and I always wondered which would be the more interesting opponent for a human over time. I don't know if its author has continued to work on it.
...
...

Thanks for reading. I'll check in again on this forum from time to time. Or you can e-mail me directly, if you wish, by clicking Support:

http://www.probe.imersatz.com/


Hello Imer,

I've enjoyed playing Probe a lot in the past couple years. I find it a friendly-to-the-user game and I have many setups saved and many games recorded there. I particularly enjoy the Milton Bradley pieces on Probe that I grew up with. (The colonel is a colonel on Probe and NOT a major! lol) And I am happy to read about the learning aspects of Probe. I, too, will look for that setting to turn on. I do wish Probe played on Android so would agree with that added feature!

As for Master of the Flag (which does play on Android, by the way) I have a question for you. Earlier in this same thread I wrote:

MF does not allow you to save setups whereas Probe does. And the game aids are either on or off with MF. When on you see the pieces marked with a dot that have moved and actually numbered if you have struck them before. In other words they are always revealed once discovered. But if you turn off the aids in MF you'll never be told what the rank of the opponent's piece is that beats you. And that's a problem with MF.


I know you are not the creator of that game but how does Probe play against Master of the Flag with the game aids situation like I've described? One way is a crutch; the other, it takes out your legs by denying critical information.

Am I missing something with Master of the Flag settings?

Thank you for your contribution to this topic of Probe!

For your information elsewhere the creator of Master of the Flag, jayoogee, did actually post a single post on this site to state that MotF did not cheat. You can find that here:
http://forum.strateg...ent/#entry20454
The complete GS&F Rules can be found here: http://forum.strateg...rum-rules-2016/

Draw Refusal Rules, specifically, can be read here: http://forum.strateg...604#entry339604

#50 Imer

Imer

    New Recruit

  • Members
  • Pip
  • 9 posts
  • Coat of arms
  • Bronze Spy

Posted 24 January 2016 - 07:26 PM

Huh, I guess I can't reply to specific posts? Oh, well, this a reply to chickenofdoom's question about enabling setup matching:

 

Launch Probe

Tools/Explore Data Files

Edit ProbeAI.txt

Change the third entry to Setup Matching=Enabled

 

For the logged on user (you on your local machine) Probe will remember whatever it legitimately learns about the original position of your pieces. Then, in subsequent games, it will attempt to match what it learns about your current setup to previous setups you've used.

 

I really did introduce this feature as a kind of lark, and it's obviously easy to subvert. I'll need to confirm it actually works on the downloadable version, but it should.



#51 Imer

Imer

    New Recruit

  • Members
  • Pip
  • 9 posts
  • Coat of arms
  • Bronze Spy

Posted 24 January 2016 - 07:45 PM

Hi, Gary,

 

Yes, now I remember the post about MF cheating: it's annoying that people can make unfounded assertions so easily. I'm glad jayogee responded.

 

In terms of game memory aids, Probe has a wide range of options including, just for the hell of it, one where you don't learn the identity of the piece that just captured yours. The default is to reveal the rank momentarily and then it's up to you to remember what you've learned, just as in board play. If MF has two settings (I've never played it), in my opinion they should be: "Reveal and then conceal again" and "Reveal and leave displayed." The latter is useful for beginners with bad memories. It sounds like the options are actually "Reveal and leave displayed" and "Don't reveal." Is that correct? If so, jayogee has omitted the board play default, which would be strange. Humans should be able to play Computer AIs the way they are used to playing each other, and computer AIs should have the advantage over humans of perfect memory.

 

When computers play against each other they always have perfect memory. The interface for computer play doesn't support a "Don't reveal" option because there's no reason to.



#52 chickenofdoom

chickenofdoom

    Spy

  • Members
  • PipPip
  • 10 posts
  • Coat of arms
  • Bronze General

Posted 24 January 2016 - 07:51 PM

Thanks for answering my question Imer. I've really enjoyed playing against Probe.



#53 maxroelofs

maxroelofs

    Major

  • Dutch Tournament Manager
  • 1,038 posts
  • Coat of arms
  • Platinum Marshal

Posted 24 January 2016 - 09:03 PM

Hi Imer,

 

I think it was me that suggested that Probe was lottoing me. 

 

I have a setup and he lottod me the first time I used that setup, within a 50 moves (This was I think my 5th game with Probe.) He took his marshal and took 2 pieces behind the lake (good ones, not moved). I understand that this is incorrect by your theory, but it is my experience. I can try to caption it on film for you if you want?

 

Greetings,

Max


To watch stratego videos: https://www.youtube....HOHXWONQMsVcOLA

#54 Imer

Imer

    New Recruit

  • Members
  • Pip
  • 9 posts
  • Coat of arms
  • Bronze Spy

Posted 24 January 2016 - 09:25 PM

Greetings, maxroelofs!

 

I'm not sure how you can capture the game, since there's no way to replicate the same sequence of moves, is there? It's possible Probe had seen enough of your Bombs to be willing to take the chance, or thought its Flag was in immediate danger or needed to go to the bathroom and wanted to end the game quickly, who can say for sure? My experience with human lottoing is intentionally high risk/high reward play where the opponent really doesn't care if he loses. Probe doesn't do that. However, it can detect if a human is lottoing and take appropriate measure, which is to say, adjust its expectations for how much risk it believes its opponent will accept.

 

I'd say that, in general, if someone decides MF is cheating or Probe is lottoing or whatever it at least means these AIs are acting unpredictably, which is a good thing provided that, over many games, these kinds of moves improve their odds of winning. Otherwise, you get counter complaints about patterns that become too obvious over time. This kind of stuff makes Stratego vastly more difficult to program than chess just to reach a credible level of play, much less ever to hope to regularly defeat top quality opponents.



#55 GaryLShelton

GaryLShelton

    Marshal

  • Moderators
  • 4,265 posts
  • Coat of arms
  • Gold Captain

Posted 28 January 2016 - 01:14 AM

Hi, Gary,

 

If MF has two settings (I've never played it), in my opinion they should be: "Reveal and then conceal again" and "Reveal and leave displayed." The latter is useful for beginners with bad memories. It sounds like the options are actually "Reveal and leave displayed" and "Don't reveal." Is that correct? If so, jayogee has omitted the board play default, which would be strange. 

 

 

Yes, you understand the MotF situation correctly.  It is a strange thing to play, unless one likes piece reveal.  There is no "normal" way to play MotF.   

 

Since you said you have never played it, check it out here:  http://www.jayoogee....asteroftheflag/


The complete GS&F Rules can be found here: http://forum.strateg...rum-rules-2016/

Draw Refusal Rules, specifically, can be read here: http://forum.strateg...604#entry339604

#56 Moriarty

Moriarty

    Major

  • Members
  • PipPipPipPipPipPipPipPipPip
  • 1,178 posts
  • Coat of arms
  • Gold Colonel

Posted 01 March 2016 - 07:55 PM

Hey, texaspete09, thanks for telling me about Master of the Flag.

I just played my first game of it. I must say, I thought it was pretty tough. I'd put this higher than Probe for difficulty but not nearly so friendly to play. MF does not allow you to save setups whereas Probe does. And the game aids are either on or off with MF. When on you see the pieces marked with a dot that have moved and actually numbered if you have struck them before. In other words they are always revealed once discovered. But if you turn off the aids in MF you'll never be told what the rank of the opponent's piece is that beats you. And that's a problem with MF.

Still, I found the play to be credible. I won but played a decent game and didn't make mistakes. I felt MF didn't make mistakes either. At least in this first go around. If everyone thinks Probe is around. 280 ELO, then I put MF at 380-420.

I just played my first full game against Master of the Flag.

 

To be honest, it is very weak. I would rate it Bronze Miner or Bronze Sergeant. All you need to do is attack. Taking captains with majors, then you know a colonel or better. Once enough info is revealed, you can totally lotto the setup and the bomb placements are predictable, flag is predictable, etc. Even if you are down 2 pieces but have your marshal it is extremely easy to capture those pieces back by the two square rule. The computer has no understanding of the two square rule. My marshal and general trapped his general and colonel easily because of the two square rule. It almost never attacks unmoved pieces with pieces higher than a captain.

 

It has no bluff, so anything chasing your known major has to be colonel or better.

I have to try probe sometime, but it doesnt download in my computer. My computer is older than I am (14+ I believe) and it cant download alot of files like exe and crx. I was surprised when i finally downloaded gravon


I have no idea what that thing under my avatar is. I've always liked stickers

#57 GaryLShelton

GaryLShelton

    Marshal

  • Moderators
  • 4,265 posts
  • Coat of arms
  • Gold Captain

Posted 02 March 2016 - 12:10 AM

I just played my first full game against Master of the Flag.


How did you play, Moriarty? It isn't a game you can normally play in my experience. Did you play with the pieces striking and revealing temporarily, or completely not revealing if they strike you? See the post by tobermoryx which explains this well: http://forum.strateg...7929#entry47929
The complete GS&F Rules can be found here: http://forum.strateg...rum-rules-2016/

Draw Refusal Rules, specifically, can be read here: http://forum.strateg...604#entry339604

#58 Moriarty

Moriarty

    Major

  • Members
  • PipPipPipPipPipPipPipPipPip
  • 1,178 posts
  • Coat of arms
  • Gold Colonel

Posted 02 March 2016 - 07:30 PM

How did you play, Moriarty? It isn't a game you can normally play in my experience. Did you play with the pieces striking and revealing temporarily, or completely not revealing if they strike you? See the post by tobermoryx which explains this well: http://forum.strateg...7929#entry47929

I played with visual aids (default) and when they strike they reveal permamently

 

its really easy IMO. Gotta try probe sometime

 

The stratego.com singleplayer is less than 100 ELO. It cant even hit the marshal with the spy. Its sad


Edited by Moriarty, 02 March 2016 - 07:34 PM.

I have no idea what that thing under my avatar is. I've always liked stickers

#59 GaryLShelton

GaryLShelton

    Marshal

  • Moderators
  • 4,265 posts
  • Coat of arms
  • Gold Captain

Posted 03 March 2016 - 03:12 AM

That's what I thought.. There's no way to compare Master of the Flag because it doesn't offer the proper setting on information. That's a shame. It otherwise might be interesting.
The complete GS&F Rules can be found here: http://forum.strateg...rum-rules-2016/

Draw Refusal Rules, specifically, can be read here: http://forum.strateg...604#entry339604

#60 Moriarty

Moriarty

    Major

  • Members
  • PipPipPipPipPipPipPipPipPip
  • 1,178 posts
  • Coat of arms
  • Gold Colonel

Posted 03 March 2016 - 06:17 PM

That's what I thought.. There's no way to compare Master of the Flag because it doesn't offer the proper setting on information. That's a shame. It otherwise might be interesting.

How come?

 

It is only fair that the human playing gets to have visual aids when it is impossible for the computer to lose memory of your pieces.

 

 

My point was that Master Flag is around 150-200 ELO, thats all


I have no idea what that thing under my avatar is. I've always liked stickers




1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users