EPT![[image loading]](http://i.imgur.com/G4E5h.png)
edit: I is dumb
[Mahjong]Tenhou Thread - Page 16
| Forum Index > General Games |
|
JSH
United States4109 Posts
edit: I is dumb | ||
|
spinesheath
Germany8679 Posts
![[image loading]](http://i.imgur.com/lo4Tc.jpg) Yeah I picked one of those hands with a LOT of possible discards. Whenever you have to discard something, you can now look at an Uke-Ire analysis of all discards that don't increase your Shanten. Would be easy to include Shanten increasing discards too, but then the amount of data becomes retarded... If you guys think it's worth having, I should easily be able add a toggle for displaying all/only good discards. ReplayAnalyzer: see post on next page for new version I also did some refactoring, hopefully I didn't reintroduce old bugs... That 2000+ lines function for analyzing the hand sure needed it. And the remainder of it still needs it. | ||
|
Rhaegar99
Australia1190 Posts
On October 09 2012 02:16 JSH wrote: This is where it got hazy edit: I is dumb This is one of those tricky spots where both decisions are not bad. You cant hold onto the 8p else you'll either be furiten from the 5p or 7p, or you'll be waiting on the last 8p to complete your pair. With the 6p discard from west, 6p and 9p are now (very) safe tiles to discard meaning that if east is bailing, they will come out sooner rather than later, giving you less outs to win. Because of this, it can makes your pin tiles hard to complete. You should be able to know that you will almost never come last. West almost has to riichi to have a decent chance to come 3rd and since hes not, its easy to say hes not in tenpai. South can't have a baiman since he discarded the dora 4p and he riichi while only 2100 points behind, indicating he has a shitty hand, either 1 han or non pinfu. No real tells on east but he needs a haneman to drop you which is highly unlikely without a riichi. Another plus to push is that west must push, making it a bit easier to complete your hand. For now I'll probably discard the 8p first since there are also 3 blockers on both 7p and 8p, and wait for the next draw. | ||
|
Rhaegar99
Australia1190 Posts
On October 09 2012 06:45 spinesheath wrote: Wow, this went way smoother than expected: Yeah I picked one of those hands with a LOT of possible discards. Whenever you have to discard something, you can now look at an Uke-Ire analysis of all discards that don't increase your Shanten. Would be easy to include Shanten increasing discards too, but then the amount of data becomes retarded... If you guys think it's worth having, I should easily be able add a toggle for displaying all/only good discards. ReplayAnalyzer: http://www.mediafire.com/?lrca64m5qbsupmo I also did some refactoring, hopefully I didn't reintroduce old bugs... That 2000+ lines function for analyzing the hand sure needed it. And the remainder of it still needs it. Your program looks pretty damn good. Remember you can't always rely on just shanten. This example you have is a good spot for shooting for a bluff half flush even though a yakuhai would be faster. Not sure if the bluff will work on the lower lobbies though. You know what would be awesome though is instead of it analyzing replays, have it such that you can input tiles manually so that it can be used during a game. It'll probably considered cheating but if your not that great with tile efficiency, a program like that may be able to help you learn faster. | ||
|
spinesheath
Germany8679 Posts
Well of course shanten and more importantly uke-ire isn't everything. After all we're shooting for highest expected value (at least in the first few hands of a hanchan), and that adds the value of a hand into the equation. Still, shanten/uke-ire is something that you can analyze clearly and correctly, while expected value depends on a lot of estimations like the time you have to fiddle with a hand before you have to bail or someone else wins etc. Uke-ire obviously also depends on the tiles that you expect someone else to be holding. Still, this is some precise information that is important whenever you are pushing a hand. Sometimes the right discard (for pure speed) isn't all that obvious too, so checking your replays can show you where your discards were bad. Watching a replay like this certainly is a better learning experience than just watching a replay on the Tenhou client. To use it during a game, having to input the tiles manually is too slow, kinda. There also is Tenhou's uke-ire calc at http://tenhou.net/2/?q=468m1156778p345s1p where you can input any hand you like and it will do the same thing I do with replays. Except for considering visible tiles though. Technically, I should be able to write a Tenhou client which works like the online client or the windows client, but ALSO does analysis on the fly which helps you out. I already discussed this a bit with a friend actually. However, I personally do consider it cheating and I am not one to cheat or help people cheat in a game against human opponents. I still can't say I discarded the idea completely. Anyways, this would be a seperate project. I could reuse lots of code so it shouldn't be too hard, but it would have to wait for a while. What I can and most likely will do is add a standalone uke-ire calc to the ReplayAnalyzer. It's still kind of like cheating, but at least you're dropping your attention to insert the data, and you won't get the visible tiles subtracted automatically. Just the raw data for your current hand. Just to make sure: I am Furiten even if the tiles I discarded wouldn't give me a Yaku, right? They only need to complete my 4 melds + 1 pair? For example: Discarded 1m Hand 23m 234 666p 345s 77s I can't Ron a 4m for Tanyao then, even though the 1m wouldn't give me the Yaku? | ||
|
JSH
United States4109 Posts
![[image loading]](http://i.imgur.com/lklK0.png) 5 renchan~ lol | ||
|
Yorbon
Netherlands4272 Posts
I'm just getting into mahjong. I think i know the basic rules a bit and now i am trying to find an easy way to practice. I was thinking about an offline game against ai or something like that to start. Is that a good start-up plan? Or are there other things recommendable? I'd love to hear some advice on how to start playing. Thanks in advance.  btw, i became interested in it via the anime akagi; part of what i know also comes from there. Is there anything about the anime i need to know with regard to normal gameplay? (besides that there are different variants of mahjong and not to bet my own life..) | ||
|
JSH
United States4109 Posts
AI is good to start with if you're still having trouble with the basic rules (ie Yaku) Try to focus on making hands and making the most efficient discards Play on Tenhou, create an account and try to level it up for a goal Try to reach 1Dan And always have Yaku~ | ||
|
spinesheath
Germany8679 Posts
before I went on to playing on Tenhou. I think it's a good idea until you get a firm grip on the rules. Now those bots are still likely to murder you if you are a beginner. Go to http://www.osamuko.com/list-of-mahjong-articles-in-english-by-xkime/ and read up on Tile Efficiency, and the top block of links titled ReachMahjong. Here is some more information on both offense and defense: http://justanotherjapanesemahjongblog.blogspot.de/ Tile efficiency is pretty much the most important skill to learn early on. Later on you can sacrifice efficiency for other causes, but you better be well aware of how much you sacrifice and if it's worth it. | ||
|
Wesso
Netherlands1245 Posts
On October 11 2012 16:02 spinesheath wrote: I played a lot vs bots here: http://www.gamedesign.jp/flash/mahjong/mahjong_e.html before I went on to playing on Tenhou. I think it's a good idea until you get a firm grip on the rules. Now those bots are still likely to murder you if you are a beginner. Go to http://www.osamuko.com/list-of-mahjong-articles-in-english-by-xkime/ and read up on Tile Efficiency, and the top block of links titled ReachMahjong. Here is some more information on both offense and defense: http://justanotherjapanesemahjongblog.blogspot.de/ Tile efficiency is pretty much the most important skill to learn early on. Later on you can sacrifice efficiency for other causes, but you better be well aware of how much you sacrifice and if it's worth it. Those bots are killing me, Thirteen Orphans within the first 8 games.. | ||
|
Yorbon
Netherlands4272 Posts
On October 11 2012 16:02 spinesheath wrote: Thanks for the tips! I played a lot vs bots here: http://www.gamedesign.jp/flash/mahjong/mahjong_e.html before I went on to playing on Tenhou. I think it's a good idea until you get a firm grip on the rules. Now those bots are still likely to murder you if you are a beginner. Go to http://www.osamuko.com/list-of-mahjong-articles-in-english-by-xkime/ and read up on Tile Efficiency, and the top block of links titled ReachMahjong. Here is some more information on both offense and defense: http://justanotherjapanesemahjongblog.blogspot.de/ Tile efficiency is pretty much the most important skill to learn early on. Later on you can sacrifice efficiency for other causes, but you better be well aware of how much you sacrifice and if it's worth it. | ||
|
JSH
United States4109 Posts
![[image loading]](http://i.imgur.com/uNHsR.png) Should I pass or take the guaranteed second It's a pretty good wait, so I have a pretty good chance of winning again, perhaps to 1st or 2nd place But I would have to win again, in the next two rounds I still have my dealer turn left as last round But hmm I ended up just taking the second, but what would you guys do? | ||
|
Caller
Poland8075 Posts
| ||
|
Rhaegar99
Australia1190 Posts
| ||
|
spinesheath
Germany8679 Posts
Yeah, seems fine on Opera... Man what the hell is this. Closed Chinitsu, pretty much at least one enemy gets one every single Hanchan. Given that the it statistically only happens in less than 1% of the hands...? I really wish my luck was a bit more evenly distributed... ![[image loading]](http://i.imgur.com/89Ydc.jpg) I got like no Ura Dora at all in the last dozen games, and then I get 5 when I didn't even need a single one? Seriously. I have now officially lost count of the number of Kokushi Musou I have attempted in the last couple of days. At least 4. 9 or 10 tiles with starting hand each time. Of course none of my attempts was successful. | ||
|
Rhaegar99
Australia1190 Posts
everythings all good | ||
|
Yorbon
Netherlands4272 Posts
First i read the blogs by xkime as recommended by spinesheath, without really getting it. I just grinded a lot of games against bots, with very volatile results. I read the blogs again with some game experience and it seemed to be a lot more understandable than first. After trying again, my 2nd try against the bots gave this result, which makes me rather proud, if i may say so. http://imgur.com/P4Url ( <-- copy paste link) I still don't care much about the discard tiles of my opponents, i'm still too busy with my own hand. ): My first priority atm is tile efficiency, building a basic hand and trying too estimate whether to call on a discard or not. I hope to keep this up :D edit: oops, the image seems to be too large. | ||
|
spinesheath
Germany8679 Posts
On October 18 2012 00:27 Yorbon wrote: I still don't care much about the discard tiles of my opponents, i'm still too busy with my own hand. ): My first priority atm is tile efficiency, building a basic hand and trying too estimate whether to call on a discard or not. That's fine, it takes a while to get the hang of analyzing your own hand. Especially when you get these huge blobs of tiles of a single suit it can get pretty darn confusing. By the way, how does everyone here count the Shanten of their hands (4 meld 1 pair hands only, chiitoitsu + kokushi is obvious)? Since I've been developing this Algorithm for Shanten/Uke-Ire, I've approached it from a couple of different angles (I still want to improve the speed of the Algorithm, lots of redundant stuff going on...). For me the easiest method I have found is: 8 - (2 * called melds) - (2 * closed melds) - (taatsu + toitsu) melds + taatsu can't be greater than 4 and with toitsu it can be 5 at most. Another approach was to simply count the useless tiles, but that confuses me all the time for some reason. Sadly you can't be greedy with the closed melds: 245568m159p1589s If you use the Shuntsu 456, it's a 5 Shanten hand. If you split it 24 55 68 it's 4 Shanten instead... | ||
|
Rhaegar99
Australia1190 Posts
As for an algorithm I think it'll be pretty dam hard. Correct me if Im wrong but even the method you havnt dosnt work for all cases. Like say 2468m2468p2468s + a honor. Thats 4shanten whereas your algorithm will give you 2shanten. Ukeire is more important though so I would say to focus on that instead. | ||
|
spinesheath
Germany8679 Posts
On October 18 2012 16:07 Rhaegar99 wrote: mmm I dont usually count shantens when I play lol. There really isnt a need or isnt even important until someone riichis, and by that time your either close so its quite simple to count, or too far away so theres no point in counting. As for an algorithm I think it'll be pretty dam hard. Correct me if Im wrong but even the method you havnt dosnt work for all cases. Like say 2468m2468p2468s + a honor. Thats 4shanten whereas your algorithm will give you 2shanten. Ukeire is more important though so I would say to focus on that instead. Nope, my algorithm yields 4 shanten for that too, notice the "melds + taatsu <= 4" part. It's 6 taatsu, but only 4 of them are counted, so I end up with 8 - 4 = 4. Uke-Ire is defined as tiles which can be used to improve the Shanten of your hand. The Uke-Ire algorithm basically calls the Shanten algorithm up to 34 times, once for each type of tile you could draw. Because the Shanten algorithm also works on 14 tile hands (Shanten = -1 in case of a winning hand, ignoring Yaku of course), I don't have to select a discard either. My Algorthms work, except for a few bugs here and there most likely (I have hints on at least 2 bugs). I just don't like the speed yet because I want to do deeper analysis (more than 1 turn ahead) as well as analysis of a big number of hands at once. It's not exactly slow, but it's not instant, so it's gonna add up fast. I do usually count shanten (well only recently). If it's too high (and the value of the hand doesn't look like it's gonna make up for it) I'm not gonna bother with the hand and prepare for Betaori. It also helps decide between Chiitoitsu and regular hands. | ||
| ||
