A 1000 game statistical analysis

[or2az]

I have just completed my 1000th game (10,936 hands) on the Pro Mahjong Kiwame app over the last 6 months. Yes, I am addicted, but I play a lot during television commercials. Here are some statistical observations and comments, which may or may not be of some interest. You decide.
Of the 27 yaku listed on Barticles yaku summary page (p.70) in order of frequency of occurrence, I was not able to get 3 kongs (San Kantsu) or a Twice Pure Double Chow (Ryanpeko). All Terminals and Honors (Honroto) and a Triple Pong (San Shoku Doko) occurred only once. Little Three Dragons (Sho San Gen) happened twice and I was able to Rob A Kong (Chankan) twice. As it turns out, these are the bottom 6 on the list. Statistics seem to work, so far. ( I also got 2 yakuman, a Heavenly Hand (Tenho), and Parenchan, a double yakuman, for 8 dealer continuances.)
The only other yaku making less than 10 appearances were Chanta and Junchan (Mixed and Pure Outside Hands), 9 and 7 times,respectively. This places them just above the bottom six on my list, well below the positions they should be in. I guess I just don't like those terminals. Maybe I should try for these hands more often.
On the positive side, Chinitsu (Full Flush), which occurred half as many times as Honitsu (Half Flush), occurred twice as often as a Mixed Triple Chow (San Shoku Dojun), which placed it well above the position it should be in.
The weirdest thing in all of this is that Chinitsu occurred 7 times more than Chanta even though it is below it on the yaku list.
I believe this is because in a video game, the bots only fixate on their own hands, without regards to what you are doing, so if you have 9 or more of the same suit exposed, they will still discard more of them, whereas a human will not. This is great, because at 5 han exposed for a Full Flush, that's a lot of points! I posted about this back in Sept. ("How intelligent is artificial intelligence")
viewtopic.php?f=5&t=53178
Okay, now for the game itself:
One out of every 6 hands ended in a draw. ( 5 of 8 draws were tenpai )
Winning by ron occurred 3 times as much as winning by tsumo ( logical, 3 opponents, 1 wall )
Winning by tsumo occurred on average, once per game ( hence, 3 rons per game, or a total of 4 wins per game, on average)
Winning hands (tsumo/ron) occurred 36.4% of the time and I got "ronned" 10.6% of the time. ( I really hate getting "ronned" and it seems to happen more than I want it to, but 10.6% seems like a decent number. If it's not, how often should one get "ronned"? ) You take a chance every time you riichi.
I opened my hand 55% of the time which seems like a high number for a game that stresses concealment. I'm still not sure whether to keep the hand concealed more often. I will get more points on the hands I win but will I win less hands and/or less games? ( I did manage to win 564/1000 games ). I don't think I open the hand too soon but the opportunity seems to arise quite often, like when you have a big lead and want to end the game faster. This seems like a good idea. Perhaps exposing your hand is another thing that is more successful in video games than in real life.
Well, that's it. Feel free to comment and HAVE A HAPPY NEW YEAR!!

Addendum: A 4000 game posting of stats can be seen here, if interested.
viewtopic.php?f=5&t=48440#p57100

[Gnom]

Wow! 55%? Either the bots you're playing against are artificial morons or you're a mahjong genius (I think best players achieve somewhere around 35% first place). I suggest you replace these bots by real players (even online) to experience some more fighting! I don't have figures but you seem to get a lot of suit-based winning hands, as you stated that's maybe due to IA. Really, I'm sure playing online will bring you lots of fun!

[Barticle]

If you want some percentages here's the source I used for yaku/man frequency:

http://tenhou.net/sc/prof.html (originally posted here by Hotel back in 2009)

For translations use my guide or this list: http://www.reachmahjong.com/en/forum/vi ... 576#p53005

[or2az]

Thanks for the info, BART. Some of those postings from 2009 ( Hotel/yttrasil ) on game recording for analysis were pretty funny, the bantering back and forth, but it was interesting stuff. Very informative. Hope they made up.
I'm seeing that playing video games is very different than playing real people. The bots do not play defense. They can't pick up clues from discards. They don't see patterns from exposed tiles. They are computers, trying to complete their own hands as quick and efficient as possible, which gives me a big advantage, i think.
I seem to win more than average and the game is fun. I'm always trying to learn, from playing, and by reading your guide a gazillion times! Thanks.

[Barticle]

It is possible for video-games to have more advanced opponent AI. Some games make a show of letting you adjust various aspects of their play style, e.g. Custom Mahjong on the DS has about nine settings and Mahjong Dream Club (PS3) lets you specify whether the bots that fill empty seats in online matches will push or fold after someone reaches. Other games have help functions that include automatic analysis of your opponents' discards, e.g. Yakuman DS might say something like "single-suit hand with two exposed sets / all Manzu suit tiles dangerous / Honitsu trend" and indicate the relative danger of each of your possible discards while Mahjong Haoh: Dankyuu Battle 3 (PS3) has quite a sophisticated analysis of your hand spanning several pages and including safe(r) discards.

It's pretty trivial to write bot code that recognizes when the player reaches and then discard their furiten and suji tiles. It seems that game developers are either lazy or, more probably, they realize that it's in their interests to make single-player games that are more fun to play by letting you win more often - and I thank them for that.

[diddysinatra]

It's pretty trivial to write bot code that recognizes when the player reaches and then discard their furiten and suji tiles. It seems that game developers are either lazy or, more probably, they realize that it's in their interests to make single-player games that are more fun to play by letting you win more often - and I thank them for that.

Yeah, its definitely both ; its very likely that one (easier for players) is being used by the devs to justify the other (devs being lazy). Every AI should at least try to get a read on you, but how good they are at it and how often they "lose focus" and "tunnel vision" their own hands should be the part that varies.