Question on APT stats

Sean O

Not sure if this is the right place for this question, if I should move it just let me know.

I just started using Advanced Poker Trainer and I'm already in love with / addicted to the performance data it provides. Still a small sample size, so I'm holding off on any big judgments. But one trend I noticed early on is that it evaluated my pre-flop and flop play as strong, but thought I was a spewy idiot on the turn.

This morning, I ran a session with exactly the opposite results (donkey pre and flop, genius on the turn). Of course there could be a thousand good reasons for this but it made me wonder about the reliability of that analysis.

I'll try to post some hands it disagreed with separately, but my general question for other APT users is: how much do you trust these street-by-street metrics, do you believe / trust the "luck adjustment", and how do people use this info to critique their own hands? Thanks.

bigburge10

I think the calculations are pretty good. I have come across a few errors--for example, I recall calling an allin on the turn with TT, and my opponent also had TT, no flush draw, so we were chopping--but in the score section, it showed this as a negative. I've seen a few other spots too. I notified the developers, and they confirmed that these were errors and they probably fixed them by now.

Anyway, I think the scoring system they use is fairly helpful. However, I think you should strive to score higher on the later streets of the hand--this is where the high value decisions come in--ie limping with 72o doesn't cost you much, but making a good bet or call on the river for the pot is more valuable.

One thing I've noticed with the scoring--and I'm still not exactly sure what to think of it--is that your individual session score can be very different based on the cards that you're dealt. So, if you played a session where you were dealt pocket AA every hand, obviously, this hand is strong enough that basically anytime your opponent puts money into the pot, it's a good thing. Whereas, if you were dealt a bunch of shit, and you folded, then your score won't be very high since you likely didn't cause your opponents to make any mistakes. So, I think that when looking at an individual session, the scores can be biased based on the cards you're dealt.

This is an excellent training site though. I love the options of choosing which types of hands to be dealt, position, competition, etc. I'll open up Flopzilla along side to practice reading hands, play hands while looking at my opponents hold cards--just as a "what if I were so good at reading hands that I actually knew exactly what my opponent had exercise", and just trying stuff out.

Sean O

Thanks @bigburge10. I read on the website that the IQ metric is basically a comparison of what I did to what I should have done if I knew my opponent's hand. Do you have a feel for how evolved that logic is? For example, let's say I bluff at a passive nit character on the river with a busted draw but get called by a strong hand. Which of the following is closest to what the software would think?

1) you bet into a stronger hand, so you have a low IQ
2) you tried a bluff but if you knew what your opponent held you would not have bluffed, you have a low IQ but maybe not as low as #1
3) you picked a good spot to bluff but got unlucky that your opponent was at the top of his range, you have a high IQ but lousy luck.

bigburge10

Sean O wrote: »

Thanks @bigburge10. I read on the website that the IQ metric is basically a comparison of what I did to what I should have done if I knew my opponent's hand. Do you have a feel for how evolved that logic is? For example, let's say I bluff at a passive nit character on the river with a busted draw but get called by a strong hand. Which of the following is closest to what the software would think?

1) you bet into a stronger hand, so you have a low IQ
2) you tried a bluff but if you knew what your opponent held you would not have bluffed, you have a low IQ but maybe not as low as #1
3) you picked a good spot to bluff but got unlucky that your opponent was at the top of his range, you have a high IQ but lousy luck.

If I'm not mistaken, these calculations are based on #2 only. In fact, if you play some hands and look at your opponent's hole cards and basically just make moves to win max value on every hand, your Overall score will be 160--each street's score will be different--for example, since you can see your opponent's cards, it's profitable to play every single hand because you'll never make a big mistake in the hand--so in this case, your preflop score will be super low, but your flop, turn, and river will be perfect.

I don't believe these scores are considering whether the spot was a good one or not to for a particular action. It's just determining whether you forced your opponent to actually make a mistake--all based on Skylansky's Theorem of Poker.

I've emailed questions to Allen Bay and Steve Stokely (I think they run the site and do the development work) a bunch of times while using the site. They are very prompt and provide good explanations.

Have you tried the Beat the Pro Challenges yet? Those were somewhat frustrating for me since it seemed like the AI players played differently against the pros than they did against me (maybe these AI are getting too damn smart?). For example, on a few hands, I raised to $20, and the AI folds whereas they called the pro or vice versa. Seems like the AI's would play the same under similar circumstances.

Sean O

Makes sense, thanks. So I guess, if I run that bluff a statistically significant number of times, my IQ will reflect how many times my opponent made the "mistake" of folding vs how many times I made the "mistake" of bluffing into a calling hand. I still wonder whether if it factors probability in (if I bet half pot I don't need 50% folds to be profitable). I'm guessing not because that would get pretty complicated. But it sounds like the main takeaway is that any particular session score is largely driven by results but over the long run it should be reflective of skill. Just like real life :)

I haven't tried the pro challenge yet, but you triggered another question for me. I wonder if the AI is capable of adjusting to the human player? If the software assesses you to be tighter than the pro, folding to you but calling the pro might make sense. But I'm just speculating / curious.