Page 1 of 1

Analysis of a mahjong hand

Posted: Wed Oct 11, 2017 7:38 pm
by or2az
Had this multiple wait hand recently and although I didn't see all the possibilities during the game, it looked interesting enough to analyze afterwards. Game was East round only.

3-dot 3-dot 3-dot 4-dot 5-dot 5-dot 5-dot 5-bam 5-bam 5-bam 6-bam 6-bam 7-bam 8-bam

It was time to make a discard and call riichi
There were 2 choices. The 4-dot or the 6-bam .

Discarding the 6bam would leave me with a wait of 2-dot 3-dot 4-dot 5-dot 6-dot .

Discarding the 4dot would leave me with a wait of 4-bam 6-bam 7-bam 9-bam .

Coincidentally, if all are available, both have a 13 tile wait.
The 2-dot 3-dot 5-dot 6-dot all yield riichi/tanyao while the 4-dot adds San Ankou.

The 4-bam 7-bam also yield riichi/tanyao while the 6-bam 9-bam also add San Ankou. The 9-bam loses tanyao, obviously.


Since the 1st situation has 3 tiles (4dots) that yield San Ankou and the 2nd situation has 6 tiles (6bams, 9bams) that yield San Ankou, I would think that would be the way to go.

BUT WHAT IF ALL TILES ARE NOT AVAILABLE.
Suppose the 3 tile San Ankou had 12 tiles available for the win while the 6 tile San Ankou had only 8 tiles available for the win. What then?

I assume the answer will be that it depends on how badly you need the points.
With a nice lead, go for the 12 tile available win.
If playing serious catch-up, go for the 6 out of 8 San Ankou win.

But what if the game is close and you are right in the thick of It?
And what if you are sitting EAST for a possible dealer mangan.

How much risk is getting that San Ankou worth for those extra points?

Any and all comments are welcome.

Re: Analysis of a mahjong hand

Posted: Tue Oct 17, 2017 2:07 am
by Ignatius
If the player were to be me, I would had tossed away the 6-bam .

Just because the wait has more tiles to wait on.

This is, obviously, if we do not consider all the potential conditions you talked about, of course.

Re: Analysis of a mahjong hand

Posted: Tue Oct 17, 2017 12:08 pm
by Shirluban
If you discard the 6-bam you get more winning tiles and you can kong the 5-bam .
(Discarding the 4-dot would allow to kong both the 3-dot and 5-dot but since otherwise they would be winning tiles it's kinda counter-productive.)

Most of the time, when you're East you'd better go for a fast hand to keep the renchan (more hands to win).

If you desperately need points, and you're not East, you can also not riichi and go for sūankō with an extra 6-bam (but don't blame me if you lose :P ).

Re: Analysis of a mahjong hand

Posted: Tue Oct 17, 2017 3:00 pm
by Iapetus
Shirluban wrote:If you discard the 6-bam you get more winning tiles and you can kong the 5-bam .
(Discarding the 4-dot would allow to kong both the 3-dot and 5-dot but since otherwise they would be winning tiles it's kinda counter-productive.)
The hands have the same amount of winning tiles. 3-dot 3-dot 3-dot 4-dot 5-dot 5-dot 5-dot has 4x 2-dot 1x 3-dot 3x 4-dot 1x 5-dot 4x 6-dot for a total of 13. 5-bam 5-bam 5-bam 6-bam 6-bam 7-bam 8-bam has 4x 4-bam 2x 6-bam 3x 7-bam 4x 9-bam for a total of 13. Tatsumaki waits may be five-way, but they have surprisingly few winning tiles.

3334555's only advantage is the higher chance of tanyao sanankou. If one absolutely needs a massive hand, then it's a better choice (though gunning for suuankou might be even better). Otherwise, the 5556678 has one tile less for tanyao sanankou, but gains 4 tiles for non-tanyao sanankou instead. That's a great trade. It also has the advantage of being able to do two different kans (and no, the kan tiles being winning tiles for the other wait doesn't matter at all), and a 7s tsumo win gets you extra fu. In a generic situation, this hand is 100% 4-pin discard riichi.

So now that we're finally back to something or2az perfectly reasoned on his own, we can talk about ways to deal with wait/value trade-off judgements.

One possible way to compare wait choices is to multiply the scores each wait would gain with the width of that wait. Thinking quickly, it would be the simplest to assume tsumo with no ura dora. Under that assumption, the 3334555 wait's total value is

Code: Select all

3*(riichi tsumo tanyao sanankou = 8000) + 10*(riichi tsumo tanyao = 4000)  = 64000. 
For the 5556678 wait, it's

Code: Select all

2*(riichi tsumo tanyao sanankou = 8000) + 4*(riichi tsumo sanankou = 8000) + 3*(riichi tsumo tanyao 40fu = 5200) + 4*(riichi tsumo tanyao = 4000) = 79600. 
Now, as tiles disappear, you can edit the formulas accordingly. For example, if someone kanned the 4-bam , the 5556678's value would drop to 63600. In a case with similar values, it's usually better to choose the better wait, since ura-dora favors cheaper hands.

(This method is a simplification and has its inaccuracies. I haven't read any books by Japanese experts who've analyzed the statistics to compute the actual expected values.)

I've heard the statistical expected value for an extra turn as dealer is about 700 points. So as dealer, you should put a bit more value on winning, but not too much.

If it's an endgame situation, choose the wait with the best chances of getting what you need. If you're only 2000 below competition, whichever wait has more tiles left is straight up better. But if you need a mangan tsumo to make a comeback, the 5556678 tends to win. If it has to be a mangan direct hit, the 3334555 is better.

Re: Analysis of a mahjong hand

Posted: Tue Oct 17, 2017 6:55 pm
by Shirluban
I mean more different winning tiles.
With the same number of total winning tiles, it gives a slight more chances to win.

Re: Analysis of a mahjong hand

Posted: Tue Oct 17, 2017 8:06 pm
by or2az
ways to deal with wait/value trade-off judgements.
Thanks for your explanation regarding the above. Was not familiar with any of that.
Are players actually able to do all that during a game? Pretry time consuming.
I assume a tatsumaki wait is 5 consecutive (or more?) numbers in the same suit.

Oh, a typo below, the 8 should be a 10. (13 tile wait and so the math works)
CODE: SELECT ALL
3*(riichi tsumo tanyao sanankou = 8000) + 8*(riichi tsumo tanyao = 4000) = 64000.