Strategies?
- Created Tuesday, July 14th 2015 @ 23:08:17
Hello
I have to admit that I'm not a good tetris player with my little bot already playing kinda better tetris than myself. I'm wondering how you tetris pro's think how a bot should behave to win the game.
--> http://theaigames.com/competitions/ai-block-battle/games/55a49af71c687b068d9f4b35 So far I see a promissing strategy by Real_Block by first stacking high with 3 columns empty and then getting a high multiplicator removing his blocks.
-->http://theaigames.com/competitions/ai-block-battle/games/55a563af35ec1d0396b0be72 Alfy_Block however prefers going for tetris scores what also gives him some not that bad results but tetrises alone don't seem that great.
I'm thinking about a strategy involving first getting the multiplicator high and then finalizing stuff with a tetris. Howerver with just one lookahead this seems difficult. What do you guys think?
- Created Wednesday, July 15th 2015 @ 00:12:32
I Think that Real_Block shows how strong (overpowered?) the combo bonus is. My bot has a more general approach. As the point mechanism may not be final, I'm focussing on looking deeper, and taking more position characteristics into account first, then only focussing on getting the biggest combo bonus.
- Created Wednesday, July 15th 2015 @ 08:40:18
Yeah I agree that the current combo system is quite overpowered. There's no point in even trying for tetrises when Real Block will just wreck you with combos before you can even build up enough.
- Created Wednesday, July 15th 2015 @ 17:44:55
Since clearing lines is pretty much solved (even without look ahead), the combo-feature is what gives this challenge the distinct spin and requires much more original thought. I think that's a good thing and while Real currently uses this best to his advantage, there'll be other bots that counter it, for example adding lines when his stack is at the top of the box.
- Created Wednesday, July 15th 2015 @ 20:41:39
I agree with Sven; this game is called block, not tetris! IMO the combo system makes this competition interesting, stacking to hope for a tetris is easier (and more boring, all bots would be the same) than crafting a long combo streak. The latter involves also more risk, which is interesting. And the imbalance of the combo counter makes that small improvements can mean a lot, which is something I like.
- Created Wednesday, July 15th 2015 @ 20:50:43
I don't know; I think every bot's still going to be the same if all we're doing is trying to build up the highest combo streak. If combos are going to be this powerful, then I think they should buff up tetrises a bit to make them worth going for.
- Created Wednesday, July 15th 2015 @ 22:14:45
But I guess the best strategy to craft a long combo streak is to leave either the right or the left side open? This would still make all good bots pretty much the same. I'm pretty sure that existing approaches to stack compact can be used to stack in 7 columns and then reducing your stack shouldn't be the hardest task.I don't see a 'counter' to what RealBlock is doing. What Sven described with attacking the opponent when he is at the top doesn't sound like a counter strategy but more like "RealBlock 2.0". As long as nobody proves me wrong I'm assuming that the only way to beat Real_Block is by fine tuning his approach.
- Updated Wednesday, July 15th 2015 @ 22:33:19
I have to agree with Superdisk. Seems like after some time, we'll see a lot of bots doing the same 2-wide, 3-wide, etc. combo stacking.
Here's an idea I'll just throw out there: combo bonuses only apply on "back to back" 4-line clears. So the first 4-line gets 8 points. The 2nd 4-line gets 10 points (only if there were no non-4-line clears between the 2 events). The third consecutive one gets 12 points, and so on. This would be even more interesting if there were broken lines (with one random hole in them) instead of solid lines. Then we might see bots attempting to clear 4-lines and clear broken lines simultaneously. That would be a sight to see!
- Created Thursday, July 16th 2015 @ 01:15:22
Whatever point system is used we will always see bots using similar strategies. But the game should be complex enough such that the best programmer wins. Throwing away the combo system and only favour 4-line clears would mean all bots will fill all but one column and hope for the I-piece, too simple in my opinion.
Preparing for a long combo streak requires much more skill than preparing for a 4-clear. I am not against favouring multi-clears more than currently the case (so there are more viable strategies to choose from), but I would be disappointed if the combo system would be abandoned.
I agree with darthbuddha that there should be something more interesting than solid lines. One idea is to offer the player that earned 4 points to throw 4-5 solid blocks (1x1 big) at all different columns, chosen by the player, in the opponent's field (falling down from top, instead of popping up from the bottom. That could disturb the other player's strategy a lot more than a solid line.
- Created Thursday, July 16th 2015 @ 01:16:55
Completely agree about the line thing. Perhaps there could also be "powerups" to allow further meddling with the other player's board, but that might get too complicated.
- Created Thursday, July 16th 2015 @ 03:24:38
The issue with removing the solid lines is that the games could easily go forever. If you did that you would have to add another finish condition, like a finite number of rounds. I don't really like the sound of that though. I think solid lines are the better of two evils.
- Updated Thursday, July 16th 2015 @ 03:59:35
Maybe if the opponent scores again before you can clear holed line, it becomes a solid line?
Edit: I also think that the solid lines piling up keep each match uninterestingly short. It's almost like whomever's algorithm is best suited to the random selection of the match wins, leaving the matches to be really close. It's sort of a shame and doesn't showcase innovative programming.
- Updated Thursday, July 16th 2015 @ 04:44:33
"[...] falling down from top, instead of popping up from the bottom."
The top surface's amenability to future unknown pieces is a product of the player's skill. (More info.) The "falling down from the top" method really undermines that skill. In my mind, this would lower the skill ceiling rather than raise it.
"The issue with removing the solid lines is that the games could easily go forever."
As another poster pointed out on the stickied thread, one solution would be to have solid lines on a "timer." Every x turns, both players receive 1 solid line. Currently, the max number of rounds is approximately:
1 piece per turn * 2.5 pieces per cleared line * 1 points per cleared line * 4 points per sent line * 20 sent lines to finish the game = 200
Therefore, giving both players a solid line every 200/20 = 10 turns would result in the same maximum game length as how the game currently is implemented.
- Updated Thursday, July 16th 2015 @ 15:22:21
After some considerations I changed my mind. The combo system still is a nice touch, but it's unbalanced. Right now there is no incentive of exploring different strategies, it's just too strong. In order to make the challenge more interesting, there must be a variation of strategies that if executed perfectly, are equally strong.
Right now, with the 10 bottom rows prepared and with the right mix of pieces, one can almost always clear 6 consecutive lines and score 25 points (2.5 points per cleared line) compared to 2 points per cleared line for Tetrising.
To remedy this, I would
change the scoring system to 3 points per double (1.5 p/l), 6 points for a triple (2p/l) and 12 points for tetrises (3 points per line).
While the best starting strategy would still be combing, the bots would need to use Tetrises afterwards due the limited space. This could even differ per game, giving a generalized AI the advantage instead of hardcoded strategies similar to generated maps in Warlight 2.
add garbage lines with holes that can be cleared
We'd need to chose between clearing lines, preparing a combo or preparing a tetris, each with advantages and disadvantages. I think this would equalize the varied strategies and make this challenge much more interesting.
change the random generator to a a 7/14/21 bag of pieces
Remove the luck factor and still keeping it random should remove some of the variance involved with a "true" random generator
- Created Wednesday, July 22nd 2015 @ 10:11:40
no hold , and 1 next ... we can't make big combos ... so we only can build 3w ... LUKC is first (if base style is ok) ... we can't clear the attack lines , so we don't need downstack ...