Hacker News Reader

Comment by mod | Hacker News Reader

Looks like you used heads up hand strengths, and then full-table hand frequencies, and computed weighted hand strenghts based on that--which is a mistake, of course.

View on HN

Here are (correct?) preflop odds according to number of players: http://www.natesholdem.com/pre-flop-odds.php

And then, what's interesting is that with 2 players ak win at 67% and jj win at 77.5% (ak < jj)

But at 10 players, ak wins at 20% and jj wins at 19.3% (ak > jj)

Illustrating the number of players changes the hand strength

That's still not weighted for how often a player plays a given hand. It assumes the opponent will play every hand equally as often.

I'd love to do this with play-raise-fold frequencies by number of people at table. Are there relatively untainted public poker hand datasets? WSOP tournament data maybe?

There's a piece of software out there called PokerTracker that allows a user to gather hand data for real-time statistical analysis. There are users out there who have very large databases (Billions of hands) and some users who will buy/sell/exchange these databases.

You can buy huge databases from other players, yes. They're full data sets, not "tainted," and have far more hands than have ever been played in the WSOP.

It's considered bad form in the poker world, and generally against site TOS, if you're a player using the data to have a read on a player before you've ever played him.

If you're just doing research, nobody cares at all. They might even give you the datasets, if they actually believe you. Or perhaps if you let them anonymize them?

Unless I've misunderstood you, it's really not considered bad form at all. It's considered pretty standard amongst professional poker players that most everyone (at least the professionals) are all using some form of HUD (heads-up display) that will overlay stats next to each opponent's name, detailing all sorts of useful information about the player's tendencies. It's almost mandatory to use a HUD if you like to multi-table more than 4 tables at once.

Collecting and mining hands you play\see is fine but its the buying and selling of other people's datasets that is bad form / banned. In the small circles of nose-bleed games, it is resembles collusion when a team shares data to undermine a single target across a set of games.

It's still pretty standard to mine hands that you didn't personally participate in. Lots of players mine nonstop. You're right that you're probably crossing the line when you buy/share other people's datasets, as you'll get information that your opponent can't possibly get without also sharing/buying datasets. That being said, of the nose-bleed players I personally knew it was still the norm to share data with your friends. I think at that level you just have to assume that all of your previously played hands are visible to all your opponents and adjust accordingly.

I think I played a little over a million hands online and probably have a dataset with millions more on top of that. If you want it, I can probably find it and send it to you. Email is in my profile.

Could you expand on this please.

Yeah, so he's using heads-up hand strengths. So how well QQ fares vs all other hands, for instance, but only one hand at a time. (It's not QQ in a pot with 9 other hands.)

Then he computing weighted hand strengths based on frequency of opponents playing a hand. So now QQ doesn't win as often, given that the opponent doesn't play 27o very often so you don't actually get to win hands vs 27o (they don't occur).

BUT to determine how often an opponent plays a given hand, he's using the IRC poker database, which is full-table hands. So we see that people play K7 infrequently, for instance, which is true at a full ring game. But in a heads-up match, K7 is played very, very frequently. 85%+ I'd say, but easily 75%. [1]

[1] I played heads-up poker professionally for a few years. I logged millions of hands.

Hand strength isn't all that matters - you also need to account for "expected winnings". A-9 has a strong hand strength, but in a real game you'll probably win small pots with that and lose large ones. 9-10 has a middling hand strength but when you win, you'll probably win big.

That's not what this blog post was about at all, and certainly not what GP asked me to expand upon.

You're not wrong, it's just off-topic, essentially.

You could say that you need to account for talent, too--but at some point you're going beyond what can be analyzed, at least currently. NLHUHE is not a solved game.

I'm just saying, if you simply used these results to guide your bidding process, any halfway-decent player would take all your money.

These charts are nothing more than starting hand strengths. There's no mention of 'bidding process' on any street.

I'm not even sure how you could extrapolate a betting strategy from the post. It doesn't even mention money, let alone stack sizes, position, or any of the thousands of variables you might factor into an entire strategy.

The post talks about "which hand is more likely to win". All I'm saying is that you don't care about how many hands you win, you care about how much money you win. It would be like measuring programmer productivity using number of lines of code written. This is obviously not an issue for someone who already understands poker, but if you didn't and you tried to use this table to scale your bets according to your overall likelihood of winning with a particular hand, then you wouldn't do well.

I think you're in violent agreement with defen. You're probably so entrenched in Texas Hold'em thinking that it didn't even occur to you that a naive player may see this and think, "This hand is great! I'm going to bet it all!" defen is just pointing out that is a terrible idea, which I think you agree with, but it probably didn't occur to you that anyone could be so naive to even try.

I am in violent agreement with defen, however I'm not with you.

Given a very naive player with a hand they've seen on the chart is great, I'd advise them to bet it all (early/preflop) rather than try to play a hand out. I think they'll be more profitable on average, particularly against a solid player, which defen mentioned.

Reply on news.ycombinator.com