Printable View
Let's use this thread to post sets that aren't very obvious and would be interesting to run through a simulator. We will pick the top 5 or so and with the help of Nikolai (@ivan_bezdomny), get run some numbers.
All sets will be heads up and assume hero is out of position.
Here are some I thought of:
- A:c: K:c: Q:c: 2:d: 8:h:
- Q:d: Q:h: K:s: K:c: A:s:
- Q:d: Q:h: K:s: K:c: 3:s:
- J:h: J:c: J:s: K:c: K:s:
- A:d: A:c: K:d: Q:s: J:d:
- A:d: 2:d: 2:d: 2:s: 5:d:
Another hard one I recently had was Ks, Qs, Js, Qh, 9h. Set the Qs up top or go for the double straight flush draw?
I know the simulator says to put 2-pair on the bottom utg. What about in position?
Suppose our opponent sets Q:c: / 3:d: 4:h: / 7:s: 8:s: and we have 2:c: 2:d: 3:h: 3:s: K:d: on the button?
What if our opponent sets Q:c: / 2:h: 3:d: / 7:s: 8:s: and we have 2:c: 2:d: 3:h: 3:s: K:d: on the button?
Come on you guys, this is your chance to run some interesting hands through the simulator. I know everyone has some cool situations to study - let's add more hands to this thread.
utg: A:s: K:s: 2:s: 2:h: Q:h:
Time for us to give some of these to Nikolai. I'll give him some today and we can add more later.
Sent Nikolai 3 situations:
1. Ac Kc Qc 2d 8h utg (from Carl (givememyleg) in the forum thread)
If there is no flush draw then Ace mid and Queen top is strong. King would go either mid or top but I'm not sure which. In this case I'm guessing that we put the royal flush draw down bottom so that it is something like X / 2d 8h / Ac Kc Qc.
2. Our opponent sets Qc / 3d 4h / 7s 8s and we have 2c 2d 3h 3s Kd on bottom (from me in the forum thread)
I think the simulator says to never split pairs utg (they go on bottom). In this case there is only one trey left in the deck. What if it is reversed such that our opponent has 2h instead of 3d? In other words, when do we split pairs and put one of them mid?
3. We have: Qh / 5s 3d Kd / Jc 2c 4c, villain has Qs / Kh As 3s / 8c 8d 6x (we're utg on our 2nd draw. First discard was 6d. from a new bot we're looking at)
The bot was not aggressive: 4s top, Tc bottom, Qd discarded
I thought it would be better to be more aggressive: Qd top, Tc still bottom, 4s discard
Nikolai ran the sim for #3. The bot made a mistake by making the conservative play (Q4 top) instead of the gamble (QQ top).
Quote:
Abridged Sim output below. In short, Tc-bottom, 4-top comes out to EV
-2.0pts. Playing the Q up top is almost two points better at EV
-0.2pt.
It does show that the volatility of results is much higher for the
Q-top move, as you'd expect.
Interestingly enough, my AI (the quick-play version, not the full
simulator) also backs off from the gamble move here. Which makes
sense, as it's not exactly trivial to compute that going for it here
is worth it. It might not be, against a bad hand that we could scoop,
or against an opponent that's fouling.
To a human, it's easy to see that Q-up is the move here. but a bit
harder to calculate by formula.
Also, the real mistake here was probably playing K-middle earlier in
the hand. You should be entering this move with KQ on top, and 35 in
the middle. Which make going for it that much easier. (Since we can no
play a live pair in the middle next move.) Assuming that the 3 & 5
came first, I never put a King in the middle in that spot.
Nikolai ran the #2 hand and it is really interesting.
D. Outside the box: K-top, 2233 middle!Code:Opponent hand: Qc/3d 4h/7s 8s
My hand: 2c 2d 3h 3s Kd
Card to play: above
Known dead cards: just the 5 our opponent is showing above
We'd like to know our best top/middle/bottom set.
For sim sample size 10601: Mean: 1.83
A. Default K-top 2233 bottom.
For sim sample size 7101: Mean: 3.06
B. Perhaps K-middle (2233 bottom) is better here? Since we less likely
to fill up, and can’t make middle two pair.
For sim sample size 5901: Mean: 4.10
C. Split pairs. K/33/22 bottom (since Deuce are 100% live)Quote:
Much better than K-top. Which makes sense, right? KK middle here would
be great. Easy free roll for QQ top, or JJ, TT, etc.
For sim sample size 5401: Mean: 3.93
Here is what Nikolai said in summary:Quote:
This matches K-middle, though does not exceed it.
Quote:
I. Despite what looks like good hand for your opponent (Q-up, 78ss
back), your hand is at significant advantage, no matter how you play
it.
II. Splitting the pair is fine, and comes out to same value as
(empty)/K/two pair. This is because both are good moves.
It's easy to under-estimate the value of two pair in the back, and
this demonstrates why. Even a low, semi-dead two pair is +4.0pt
against what looks like a solid opening set by your opponent.
Here, the hand you want to make is really clear. Two pair back, AA or
KK middle, and QQ, JJ, TT up top. That's a nice hand.... and the bonus
kicker of the back full house, which can still come it.
Under very favorable circumstance (K beats Q on top, etc), splitting
the pair is fine here. But's not generally a strategy to look for. Not
because it leads to a bad hand... but because even a low, semi-dead
two pair in the back is a very good hand.
A:h: J:c: T:d: 6:c: 6:d: utg might be good to see
Nikolai ran Carl's hand and the 3 to the royal flush on bottom set beats the KQ / A / X type fl set.
It is an interesting tradeoff. I like the fact that the simulator shows why the extra fl opportunity of KQ / A / X isn't worth it because of the increased foul percentage.
Opponent hand: unknown because we're setting utg.
My hand: A:c: K:c: Q:c: 2:d: 8:h:
Card to play: above
Known dead cards: none
Quote:
The answer is pretty obvious. No reason not to just play A:c: K:c: Q:c: on the bottom and 82 off suit in the middle.
However, with my new breakdown, it’s more interesting to see why.
p1=/2:d:8:h:/A:c:K:c:Q:c:
…
For sim sample size 3601: Mean: 1.50 [1.09, 1.90]; STDEV: 12.34
For sim sample size 3601: Foul: 13.6% ptsNoFoul: 3.3 gotoFL: 7.7% Scoop: 31.8% Scoop'd: 20.4% BothFoul: 1.9%
Time spent: 0:04:39:34.83
…
This move only goes to FL 8% of the time, but it also only fouls 14%, while scooping 32% of the time, and winning, on average +3.3 points when it doesn’t foul. Not an amazing spot, but solidly above-average move.
Now compare this to the second-best move, which aggressively goes for Fantasyland.
p1=K:c:Q:c:/A:c:2:d:/8:h:
…
For sim sample size 801: Mean: -0.31 [-1.25, 0.63]; STDEV: 13.57
For sim sample size 801: Foul: 33.8% ptsNoFoul: 3.9 gotoFL: 21.2% Scoop: 22.0% Scoop'd: 35.3% BothFoul: 7.7%
Time spent: 0:00:59:10.37
…
I didn’t need to run it long to see the difference. You’re fouling 34% of the time, which is pretty high. For that price, you go to FL 21% of the time… which averages out to a slightly better hand when you make it. But you are not adequately compensated for doubling your foul rate.
I’m tracking a lot more numbers alongside, but only outputting a few at the end, so as not to confuse myself too much. Some of these stats are a lot more clear, for later in the hand spots. Where you’re more directly trading off Fantasyland, scooping, getting scooped, and foul risk.
Hope this is interesting.
I like the way the stats are presented, but it could use a note as to the Confidence Interval (CI) in use that is implied by the [min max] after the mean. The CI is a vital piece of info which tells you the "weight" on an error range; the error bars are meaningless without a CI.
He's given us enough info to sus it out. It's a 95% CI. (using a normal dist. w/ mean, stdev and number of trials given)
This guy seems to be my kind of guy. :p
***
I wonder how the bot is playing as pertains to early draws and setting a flush vs waiting for a royal.
Does the bot ever abandon a flush draw for a pair based hand?
If so, what drives that decision?
Good questions, I don't know.Quote:
Originally Posted by MadMojoMonkey