|
Post by Rycchus on Oct 7, 2007 11:03:01 GMT -5
dni, I think you read Taliesin wrong. (Tali, you should put a "Nothing personal" sig in your profile People keep taking offense ) I don't think he was arguing in order to be competitive but because the situation required it. Sometimes there are cases where the issues are more black/white than others. If I went round telling people that S/F was the best opening strategy, sooner or later someone would come along and tell me that I was wrong, and why I was wrong. The point is that when someone argues with you and it's clearly mainly a matter of opinion then you're fine to keep at it, but if it's a matter of fact, one of you is likely to be in the wrong, and if it's you you need to realise that and accept it. I've said some stupid stuff on here before, and when someone points it out you have to accept it - otherwise how will you learn anything? Not that I'm implying you are/were being stupid by any means - I'll leave the harsher adjectives attached to me alone But seriously, accepting the other person's opinion, listening to it and maybe adjusting your own are the best way to have a "flexible" discussion - which you can't tell the other person off for not having if you're not doing it yourself either I hope you're not too upset by the crossed wires and confused signals that have been floating around. I don't think anyone meant anything negatively. It's good to have your voice around the forum; I think I've learnt a lot just from reading this discussion. Let's not see you go running off, now
|
|
|
Post by dni on Oct 7, 2007 13:43:15 GMT -5
2 Slartucker: Oh, i feel that i got used to hearing this words... Complete misunderstanding. Maybe. So i don't want to make him feeling bad in any cases (whether it causes of my opinion or long conversation). He wrote quite clear. So let it be. I don't sense what you mean with the 'truth' in the AI-player method. Which of truthes? Undoubtedly. I can write on russian to be exact in my words. But i don't think that it clears the situation. By the way, i understand what you said. It was really hard to me to move out the black/white perception of world, and i don't have much of wishes to chain myself to this procrustean bed even i'll get ultimate words of truth then. 2 Rycchus: Hih, yet another 'wrong'. It's a interesting perspective... However, i don't read it. I know and feel what Taliesin wants to say and why he wants to do it. I think that the better decision is shift myself to parallel dimension for him. Somekind of double-p. You feel it? What's a situation? The destruction of ever truth? I don't proclaim any kind of truth. The idea was just a non-machinery algorithm of making a decision. That's all of. Where in my words were the 'BEST' word? If i have an idea of unpredictable method of decision making (only an idea, i have no time to write even a strict description of this idea), that's all. All i said that it's much interesting that boring move calculation. The only fact of this will be a statistics of ready-to-play method. I don't try to refute neither basics of this game, nor deep aspects of it. I'm ready to accept any opinion -- but i don't see any of these except 'I know the best method and i can argued it's bestness'. I don't need a 'best' neither in warlocks, nor in real life. All of most interesting happens not at tops, not in depths. It happens in casualily. No problem. I tried to explain my position but i can't do anything if all my words heard from position 'speeder, higher, stronger'. When i said that minimax method is fully predictable it causes argued answer that no better method more. So what? I need to take this opinion? No problem. No any confusion, i was cast amnesia to myself to eliminate confusion. Thank you. I've got a large lesson of this situation. So i think i need to feel an opponent of conversation much before than making a chat in depths. No any running, but it will be wrong to restore what, clearly said before, need to be terminated.
|
|
|
Post by Rycchus on Oct 7, 2007 14:26:22 GMT -5
Hih, yet another 'wrong'. It's a interesting perspective... However, i don't read it. I know and feel what Taliesin wants to say and why he wants to do it. I think that the better decision is shift myself to parallel dimension for him. Somekind of double-p. By "read" I meant "interpret". So, yeah, what you "felt" was misjudged (if you don't like "wrong" You feel it? What's a situation? The destruction of ever truth? I don't proclaim any kind of truth. The idea was just a non-machinery algorithm of making a decision. That's all of. Okay, I'm struggling a bit with your English here. I don't know what you mean. I don't think either of you were heading for the destruction of truth. Where in my words were the 'BEST' word? If i have an idea of unpredictable method of decision making (only an idea, i have no time to write even a strict description of this idea), that's all. All i said that it's much interesting that boring move calculation. I didn't say you said "best". I was using a parallel example. But still, you did say things like It's much smarter to examine player's habits and play against that player. and if player will change his strategy and form of tactics from one game to another to fake the auto-teaching methods. It will be too vulnerable to player 'cause he can't keep in mind all the forms of game in one time...which do suggest you're trying to make an argument with which method works better. Perhaps just an emphasis throughout all of what you were saying that you weren't interested in building an AI that could win, but one that would be more fun, could have avoided a lot of the "arguing" The only fact of this will be a statistics of ready-to-play method. I don't try to refute neither basics of this game, nor deep aspects of it. I'm ready to accept any opinion -- but i don't see any of these except 'I know the best method and i can argued it's bestness'. I don't need a 'best' neither in warlocks, nor in real life. All of most interesting happens not at tops, not in depths. It happens in casualily. My point was kinda that by now most of the top players have played enough and learnt enough that their "opinions" on the fundamentals of the game (like minimaxing) can be taken as fact. Some of the stuff will be opinion, but there's a large enough database of experience between them ("statistics of ready-to-play method") that a lot of it is more than that. No problem. I tried to explain my position but i can't do anything if all my words heard from position 'speeder, higher, stronger'. When i said that minimax method is fully predictable it causes argued answer that no better method more. So what? I need to take this opinion? No problem. You're right. It is predictable. I see your point on this one. And this is why the top players, when they're playing each other, will occasionally gamble on things - otherwise their opponent could take advantage of them. But IMO your AI would need to at the least first minimax and then perhaps add a slight random chance that (if the consequences not too severe) that it didn't act that way. But really you probably know more than me about AIing and programming of any description. I know the fundamentals of the theory but nothing at all of the practical... Anyway, like Slarty said, I think a lot of this is language issues and problems in communication... Don't take anything I say too seriously
|
|
|
Post by dni on Oct 7, 2007 15:11:38 GMT -5
2 Rycchus: Hope so. Oh i know... my bad english... I'll try to explain... You said 'because the situation required it' but what i think of 'requireness', is a stand for some kind of 'truth' (correct me if i was wrong). So i don't see any great point which quietly need for stand for. The strategy of game? It don't even touched. The tactics of game? Maybe, the minimax is panacea i feel. Valuation methods? Not touched. Methods of deep search cutting? Not touched. So, the conversation is stopped on point, defending of which is not so valuable. However, if it is valuable, AI never be written. Yes, it was. The AI can predict probabilities of player moves then. So simple valuation of position can be improved by collection of player games and database of his habits. Each player has personal patterns of perception and methods of play. I know that humans can learn faster that auto-teaching methods, but computers can collect numerous number of facts and gather statistics of play; it's what humans can do only in common. And if player wish to beat his own statistics to fake auto-teaching methods, he put all his mind forces to it and he will become quietly vulnerable. Better? Stronger? Why? For what purpose? In general, all AI-players will be for fun, don't they? And this game is for fun too, don't it? What funny in AI which will beat any human player? But if human can find a weakness in it -- not a countable weakness but feelable weakness -- it will be much more interesting for humans. P.S. I like play with humans, especially in real-time, 'cause i can use psychological influence and sometimes force to make bad move. With count-position AI it don't works -- you need to count nearest tactics exactly and win AI on strategy. OK, let it be a fact. I know why too. Maybe. I don't feel minimax for now and i think that it consumes much of time to examine. Like to play of myself inner state. From other side, position play with minimax method can be used with upper than 5 hp. If you have below than 3 hp, all may be lost in two-three moves. So i don't know what is minimax with 1 hp at all. Programming? Maybe. AIing? Maybe not. It's sad that i can't insert rightful translator in my head.
|
|
|
Post by Rycchus on Oct 7, 2007 17:33:09 GMT -5
Oh i know... my bad english... I'll try to explain... You said 'because the situation required it' but what i think of 'requireness', is a stand for some kind of 'truth' (correct me if i was wrong). So i don't see any great point which quietly need for stand for. The strategy of game? It don't even touched. The tactics of game? Maybe, the minimax is panacea i feel. Valuation methods? Not touched. Methods of deep search cutting? Not touched. So, the conversation is stopped on point, defending of which is not so valuable. However, if it is valuable, AI never be written. I'm not sure what you mean about it being the panacea. Is this bad? Are you saying it's too strong? Minimaxing is just one of a range of (complementary, not mutually exclusive) tactics in a game, but still it's one you won't get far without. What do you mean by valuation methods and deep search cutting? I think to say minimaxing isn't valuable is going too far, even if in your opinion it's not the be all and end all. Don't understand the last bit either. Yes, it was. The AI can predict probabilities of player moves then. So simple valuation of position can be improved by collection of player games and database of his habits. Again, you say that you're not making claims about what's "better", but you are. You use words like 'improve' and 'smarter' and stuff... Anyway, IMO it's best to just assume that your opponent will make the best move. All top-level players will do this anyway, and therefore it's an identical process to work out the best move, rather than working out what the individual will play. As for lower-level players, well what if they improve? Your statistics of their games will take a while to reflect this and in the meantime they'll beat you. There is one situation in which this could theoretically be better than position valuation, and that's where you're down to a real 50-50. If it doesn't matter what you'll pick, you might do better knowing what your opponent "usually" plays. Better? Stronger? Why? For what purpose? Er, no, I was saying that this was what you were saying. See above. In general, all AI-players will be for fun, don't they? And this game is for fun too, don't it? What funny in AI which will beat any human player? But if human can find a weakness in it -- not a countable weakness but feelable weakness -- it will be much more interesting for humans. Yes, of course. But I imagine it would be very difficult to build an AI that was perfect simply because of the huge number of axioms you'd have to give it. Its weaknesses would come where it misjudged which position was better (and the program would then be tweaked) not because it didn't try to judge it. P.S. I like play with humans, especially in real-time, 'cause i can use psychological influence and sometimes force to make bad move. With count-position AI it don't works -- you need to count nearest tactics exactly and win AI on strategy. Well, for me, I play strategy games because I like playing by strategy There is a certain element to playing against real opponents that you'll miss out on with an AI - but I think you'd miss out on it with any AI... Maybe. I don't feel minimax for now and i think that it consumes much of time to examine. Like to play of myself inner state. From other side, position play with minimax method can be used with upper than 5 hp. If you have below than 3 hp, all may be lost in two-three moves. So i don't know what is minimax with 1 hp at all. Um... what? Minimaxing means minimising the maximum possible loss. It's applicable to any position. It's even more applicable when all may be lost in two/three moves. It means you don't choose the gesture paths that result in you losing everything in two-three moves. Of course when close to death there is sometimes more gambling involved, because often sooner or later you'll have to take a death-risk either way. It's sad that i can't insert rightful translator in my head. Yeah Maybe Russian should be the next language on my list. Of course, I've got to get through Chinese, Greek, Italian and Japanese first. (Overachiever? Me?)
|
|
|
Post by Slartucker on Oct 7, 2007 17:33:59 GMT -5
*facepalm*
Let me say this about the language issues. If you want to say "You feel it?" when a native speaker would say "Is that really what you think?" then go ahead. It sounds strange, but I was able to figure it out. However, when you start playing fast and loose with conjunctions and pronouns -- which is the case every other sentence -- it becomes very hard to know what you are saying. We can work out the terms you are using, maybe, but the relationships between them are very hard to understand. That makes any kind of precise, analytical discussion, like this one, impossible.
|
|
|
Post by Rycchus on Oct 7, 2007 17:37:23 GMT -5
Yeah. It's why I let you (dni) and Tali get on with it, before... mostly I didn't understand what you were saying.
To be honest, I'm just very confused about the whole thing. I get some of it, but not really enough to make a coherent discussion in reply : S
|
|
|
Post by Rycchus on Oct 7, 2007 17:39:09 GMT -5
Seiðmaðr?! Another crazy one! Are there a whole spiel of these with foreign letters in? *suspicious*
|
|
|
Post by Rycchus on Oct 7, 2007 17:45:56 GMT -5
Also, dude, If neither of us know anything about it, what say we go away and leave the discussion to those who do, huh?
|
|
Derfel
Ronin Warlock
Did I Do That?
Troublemaker
Posts: 283
|
Post by Derfel on Oct 7, 2007 21:50:02 GMT -5
As The Warlock Turns...
|
|
|
Post by Slartucker on Oct 7, 2007 21:53:47 GMT -5
As The Hell-Plane Turns? One Life to Live... not.
|
|
|
Post by vermont on Oct 7, 2007 22:23:31 GMT -5
|
|
taliesin
Ronin Warlock
Grand Master
Posts: 156
|
Post by taliesin on Oct 8, 2007 10:29:20 GMT -5
Okay, dni, it seems we're having some communication difficulties. Sorry for sounding harsh. I'll try and lay out what I'm saying as clearly as possible.
My background, before I begin: I have a year of AI study at the University of Edinburgh; I studied it and computer science in second year. I elected not to continue to joint honours, instead specialising in software engineering, but I do have a fair basic level of knowledge on greedy algorithms, expert systems and the usual standard AI stuff. I've written my share of Prolog, and I've some slight experience with both the theory and practice of neural nets. I don't claim to be an expert on AI, but I have some formal education on the topic.
I also have a reasonable level of skill as a Warlocks player.
Now, back to the topic at hand.
Various people including vilhazarog and nawglan have expressed interest in writing an AI for Warlocks.
It is important that we give them good information about methods of writing an AI so they do not spend a lot of time doing something that will not work.
Suggesting anything but state-space search as the basis for the AI is bad. Neural nets on their own are weak even at simple games like tic-tac-toe/noughts and crosses. They are useful only as additions to an existing game AI. It is fine to talk about it "not being the best method" - but in fact this method will not produce an AI that challenges even very new players, and it would require a lot of work to get anything at all. It is not fair to nawglan or vilhazarog or whoever else is thinking of doing this to suggest to them that this is a sensible method of writing an AI. I think it is important that we get this clear, in black and white, if you like, so that they do not waste their time. I think it is also important that they are made aware of how game AIs are written, why they are written this way, and how much work it is to write such an AI.
We are not talking politics, here. We are doing something like building a bridge. Some ways of building a bridge really are better than others, and we do not want a bridge that falls down.
Now, you criticise minimax because it is "predictable". In fact, it is not predictable unless you evaluate the positions in exactly the same way as the person you try to predict. Imagine two players whose opponent has an ogre, and they both have the chance to kill it. However, killing it will mean they get hit for two damage. One player says, "It is worse to lose damage than to let my opponent keep his ogre." The other player says, "My opponent's ogre will be trouble in the future - more trouble than taking two damage is." How will they be able to predict what each other will do?
In fact, you only need to worry about being predictable after you are up at the top, where the best players evaluate positions in quite similar ways.
Changing topic, if we consider using players' past games as something to help the AI, we find the following things are true: 1) The AI must understand the game really well, so it knows what moves are best against the move it expects the other player to play. You need a really good AI already before you can consider doing this. 2) Predicting the other player is not always useful - if you know your opponent will complete the FoD but you cannot stop it, what use is knowing? Players are most predictable when they are doing things that cannot be stopped. At other times, they are less predictable. 3) There are many unequal situations, where players play safe almost all the time to protect themselves. If you know for certain however that your opponent will make a move based on what he has seen you do in the past, you can make big gains by not playing safe. 4) Identical situations rarely come up, and players will play differently based on what monsters there are, how many health they have, if someone has a banked spell, Disease, and many more possibilities. They will also play differently over time because they learn more about the game. You need to know them from thousands of games, and they have to have stopped learning, before you can predict them accurately.
So, it might be interesting, but it sounds like it might be easy to exploit for a good player and it would need to already have a very good AI as a basis.
|
|
|
Post by dni on Oct 11, 2007 4:35:57 GMT -5
2 Rycchus: I mean that minimax becomes somekind of idol. I know that i may not argue about it -- i can evaluate my possibilities in this game clearly. But i see a force in this game so i think that it's not one possibility to play. By the way, i've get enough on this subject so i don't want to dispute about minimax no more. As i said before, let it be. And not for me is legal to argue of this. I guess it. Valuation method is 'formula' (if say simplier) that can calculate current position or any other position. Deep search cutting is a limit of direct analysis of possible moves. You cannot analyse all the game so you need to cut your search depth according to position valuation. Each of this terms are quite useful in 'calculative' methods of playing, and there are severe useful in any other methods. Ok, these words and some other words you may interpret as 'better'. But i mean 'better' not for result (winning), i mean 'better' for distancing from direct calculation (closer to using auto-teaching methods). For example, you use position analysis method as function that returning a value. The bigger value you get, the more valuable (stronger, advancer) position you have. And there's one key aspect. You need to describe this function for AI. Let's see at simpliest method: sum of weights. You need to attach a value for each element of game (for example: having a goblin -- 10 points, paralysing with disruption an opponent's hand -- 25 points, successive FoD -- 10000 points, etc). So where will you get these weights? You need to assume them. In order to use any calculative method, you need to go through thousands of variands and minds. In the other side, if you use self-teching methods, all you need is starting weights (surely, assumed too) and statistics and methods for changing them. You can mean 'best' like mathematician -- best of all possible variants of game. In this case i'm totally agree with you. But you can't get all possible variants in reality. Frequently spoken, 'best' move is a best of your ideas. This case comes to 'if i'd was an opponent, which move i'd choose?' But you are not identical with your opponent. Best of your choice is not best of opponent's choice and, thoroughly, not the best of all choices. They will improve statistics then. Sure it can be. But not for all of the time. Statistics will be shifted away according to new behaviour of player. Yes, it's one of these situations. Generally spoken, full self-teaching algorithm don't have any axioms. But in this case you need a huge time even to teach it of game rules. I think that there's need to build mixed method which can calculate moves and evaluate positions but making his move will be based on self-teaching algorithms. So this may come in any case (even in 'minimax'). Evaluation is a big problem. I'm always envy people who really can make a strategy. For me, it's a unfeelable ability. Playing for me happens only on my mood and state. I can only guess it. Uh, defensive tactics? What about maximising the minimum loss for opponent? So, if you have 1 hp, you have to 'palm' every move? Or 'clear' minimax method works only if you have more than 5 hp? Oh my god... I know a person who knows 9 languages... No more happiness she get from this... 2 Slartucker: Oh, i see and i understand what you want to say. The problem for me is be sensitive on language fine points. I assume that some of my expressions may give offensive shades, but in order to feel this i need more practice with native speakers (so i do it now?) Hmm... probably i know what you mean. I feel it in most messages there... like a person who turned his auto to the wrong lane and surprised with a fact that all on the road drives at wrong lane. In order to understand some sentences which i read there, i need to use 'heuristic' analysis to get possible meanings... By the way, last message of taliesin is understood for me much more clearly. I know the problem of 'first-form-school' language, and i thank him for that. Yes, you're right. 2 taliesin: Thank you for reply and for 'understandness' of your message. I will write you shortly 'cause i need a time to work now...
|
|
|
Post by Rycchus on Oct 11, 2007 7:00:45 GMT -5
About the 1HP thing - I explain about a sentence later that yes, you do have to gamble to some extent. Not just Ps, but you also need Ws every turn at 2 or less to prevent from WFPs. But even if you're WWPP.../PPWW...ing, you usually need to gamble at some point to get ahead. And about the languages - they're the ones on my list that I don't know yet, or only know to some basic level. And I forgot about Arabic - that should be on there too... But, y'know, I might learn Russian by the time I'm 90 or something...
|
|