Autorate rules debate for PokerTracker Omaha

[gergery]

So after around 5k hands of limit O8 mostly at 3-6, with a bunch more data mined from all levels, here are some more thoughts on auto-rating:

1. The purpose of auto-rating is to help you make a better decision. So after looking thru the different criteria, I ask myself, “What specific decision am I going to be able to make better from knowing this – and it should have a big impact on my earn rate?”

2. The criterion in my opponents I find myself looking at most often is VPIP. I want a table with at least 2 players over 40, and over 50 is better. Good tables can have 3 people over 50. Bad tables have everyone below 32 or so. Also, I want to be in the tighter half of the table, and my VPIP is about 27 right now. I think my VPIP is probably a bit too high, and I’ve been tightening up and its been dropping. My win rate is ~6BB/hr/table. Not sure if I’m just running well.

3. O8 plays looser than holdem. Datamining shows levels up to 10-20 still have several seeing flops, and when I look at the summary tab in my holdem games and my O8 games the O8 median point is several VPIP points higher than holdem, on average.

4. It seems like there are a number of hands that are basically close to EV neutral. I think you can play these or fold these. Ie. hands like A2 with nothing else going for it, or 2356. So my theory is that your first 15% or so of hands are fairly profitable, and your next 15% or so are sorta EV neutral. Not sure if this is a valid theory or not. The main reason to play those hands is so get paid off on your good hands, ie. disguise the times you are in there jamming with double nuts, and so that opponents can’t narrow your range of hands too well. After about 28%VPIP or so, you move into marginal or EV- hands.

5. If my opponents have VPIPs below 18 or so, it’s not so tough to put them on hands, and it can be profitable to bluff and raise loosely vs. them since hand values run more closely together. If they have VPIP over 30, then they’re playing weak values.

6. The above numbers are for tables where the range of VPIP is typically 15 to 35 or so. If you are at tighter tables, you must tighten up and adjust the numbers downward. However, all games I’ve mined so far from 20-40 on down are looser than I expected, averaging ~25 or so in VPIP.

7. The other important criterion I look at is Won $ at Showdown. This basically answers, “how well does my opponent evaluate their hand fit with the flop”. My number here is around 70 right now, and the average for my opponents is around 60. I still feel I’m making some mistakes, so I think you want this number higher – 75-80 seems about right. If its too high tho, then you’re not making some weaker calls that have good odds or just getting lucky in not getting counterfeited/ outdrawn. Some of my worst opponents have numbers at 50 or below.

8. Actually, I think what might be the best metric of all with regard to Won$SD might be 1) “% of the time you win half the pot or less, whether its high or low”, and 2) “% of time you win ¾ or more”. Really what you care about is – How often does my opponent play to scoop, vs. play for half?

9. When you can find an opponent with VPIP above 40 and Won at Showdown below 50 you really have some good times.

10. Went to Showdown is also somewhat helpful. I basically use to determine when I can try to push someone off their weaker hand on a T93 type flop. Also, I think it might to helpful to determine whether you should raise the turn and expect them to call vs. fold.

11. I don’t think aggression is a helpful statistic at all. PFR raise numbers are much lower at O8 than HE, with most players averaging around 3-6%. Postflop I haven’t looked at it, but calling is the correct play in O8 much more often than in HE, so I think overall aggression levels are going to be lower. Furthermore, I don’t see what it really tells you that is going to be helpful for a decision. In HE, most of the time people don’t have anything. So it helps you call down a LAG if he’s aggro. But in O8 most of the time people have the nuts or a draw to nuts. So you’re just evaluating your hand to decide whether to continue or not, and using your opponent as an adjustment to that, instead of as one of the primary decision-drivers as in holdem.

12. The biggest correlation I’ve seen so far between earn rate and another characteristic is with PFR. The high earners seem to have 6-9% or so. Not sure what to make of that yet. It seems as tho raising some preflop may be the best strategy. I’ll hypothesize that it ties people to the hand more when you have an advantage, so your good hands end up winning more, even tho your variance and bad beats will be higher. Note – this still doesn’t make PFR a helpful measure, as I’m not sure what I’d do differently if I knew this about my opponents, and even if I’d do something differently I’m not sure how big an impact it would have. Maybe occasionally there is a giant aggro LAG so I’ll check-call, check-call, check-raise him instead of betting, but that’s about it.

Any thoughts/responses to this greatly appreciated.

Note: cross-posted at Pokertracker
--Greg

[gergery]

I think those look pretty good at first glance. i'll look thru my database again to check.

I'd also take out aggression and add in Went to Showdown, but that's me.

[sammy_g]

gregery, you make some excellent points.

[ QUOTE ]

12. The biggest correlation I’ve seen so far between earn rate and another characteristic is with PFR. The high earners seem to have 6-9% or so. Not sure what to make of that yet. It seems as tho raising some preflop may be the best strategy.

[/ QUOTE ]
This one I'm not so sure about. How high is the SD for these players? I suspect that raising preflop greatly increases your variance in this game, so you'll find that the biggest winners AND biggest losers in your DB raise preflop a lot. Is this the case?

I wouldn't look at opponent win rates at all yet since you can't have enough hands for these numbers to be close.

Sam

[gergery]

[ QUOTE ]

so you'll find that the biggest winners AND biggest losers in your DB raise preflop a lot. Is this the case?

[/ QUOTE ]

No. When I sort by preflop raise, many more of the ‘green’ results pop to the top. But I am open to a number of possible hypotheses as to why their results are better, only one of which is that PFR more is better.


[ QUOTE ]

I wouldn't look at opponent win rates at all yet since you can't have enough hands for these numbers to be close.

[/ QUOTE ]

I agree. With only 5k hands the numbers are at best directional. But it is interesting so see them correlated so far.

[Yads]

[ QUOTE ]
12. The biggest correlation I’ve seen so far between earn rate and another characteristic is with PFR. The high earners seem to have 6-9% or so. Not sure what to make of that yet. It seems as tho raising some preflop may be the best strategy. I’ll hypothesize that it ties people to the hand more when you have an advantage, so your good hands end up winning more, even tho your variance and bad beats will be higher. Note – this still doesn’t make PFR a helpful measure, as I’m not sure what I’d do differently if I knew this about my opponents, and even if I’d do something differently I’m not sure how big an impact it would have. Maybe occasionally there is a giant aggro LAG so I’ll check-call, check-call, check-raise him instead of betting, but that’s about it.


[/ QUOTE ]

Aha, finally some justification [img]/images/graemlins/grin.gif[/img]

I think one of the reasons you're seeing this is because when you raise PF you get more dead money in the pot so for the times when you do get quartered you usually wind up breaking even because of the dead money.

[Yads]

[ QUOTE ]
I'd also take out aggression and add in Went to Showdown, but that's me.

[/ QUOTE ]

Maybe we can somehow combine the two, but you're right that it's not quite important anymore.

[L0QTiS]

I've developed a preliminary rating rule set for PTO.

Keep in mind, this is focused on Limit Omaha Hi/Lo hands from Party/skins, Stars and Poker Room at the Micro and Small Stakes Limits (14k hands @ .5/1 and 8k hands for both 1/2 and 2/4 combined).

The goal here is to develop a rating system which can help me make better decisions based on opponent stats.

There's quite a few gaps, but I'm hoping with a little feedback, we can fill some in.


1. Maniac
VP$IP 60+
PFR 15+

2. Fish
VP$IP 60+
W$SD 50-

3. Rock
VP$IP 15-
W$SD 80+

4. Slightly loose / Bad W$SD stats
VP$IP 26.01-39.99
W$SD 57-

5. Slightly loose / Poor W$SD stats
VP$IP 26.01-39.99
W$SD 57.01-66.99


6. Slightly loose / Good W$SD stats
VP$IP 26.01-39.99
W$SD 67+

7. Good player
VP$IP 26-
W$SD 67+


Based on 23,000 hand and about a 4100 player database and a 50 hand minimum rating criteria:
570 players rated
574 players unrated

Win/Loss %

Good (113 players)
79% winers / 21% loosers

Rocks (21)
77/23

SL-L / Good WSD (77)
72/28

SL-L / Poor WSD (75)
51/49

Maniac (19)
42/58

SL-L / Bad WSD (97)
29/71

Fish (168)
16/84

Of the qualified unrated players, the most obvious group is those players with VP$P 40+ and W$SD 50+. I haven't quite figured out what to do with these players.

Comments apprieciated

[TGoldman]

I think a lot of the strength behind the Hold'em auto-rate rules was that the most commonly used statistics such as VP$IP, PFR%, and AF converge to their "true" values relatively quickly. After as few as about 30 hands, it's possible to have a decent read on the type of player that we're up against.

For Omaha, statistics such as WtSD% and W$SD (As well as % of pot won stats such 1/4 vs. 3/4 vs. scooping) provide helpful information, but I'm concerned about the sample size required to make these numbers really relevant. These stats converge so slowly that I'm afraid without extensive data mining they would only have limited usefulness.

I'd like to propose a new factor, something like a "Fold Factor" for each street, calculated as:

Fold Factor = (Bet/Raise/Call %) / Fold %.

Basically, it would give you an idea of how often a player continues with his hand onto the next street, relative to the frequency that he folds. Since the Fold Factor would be comprehensive of every street, it would allow us to gather information every time a player sees a flop.

Clearly this stat would be very VP$IP dependent. For example, players who stick to premium starting hands are going to hit the flop enough that it's oftentimes correct for them to continue. However, I think the combination of VP$IP and a FF may start to give a decent overall profile of a particular opponent. Unfortunately, we would have to do the calculations by hand to test this hypothesis as PTO doesn't have this statistic. Is anyone willing to give it a try? Admittedly I'm new to Omaha, so there may be nuances about the game or playing styles that would render this approach meaningless.

[Yads]

L0QTiS I like your rating system, but I think your VPIP is really high for a good player. I think it should be at least below 22 and probably below 20 for a full 10 handed game. Also what about the players that have low VPIP, but don't have a high enough WSD? I personally fit into this category at the moment. Would they just be considered unrated or would they have some sort of possible tag?

[L0QTiS]

[ QUOTE ]
...what about the players that have low VPIP, but don't have a high enough WSD? I personally fit into this category at the moment. Would they just be considered unrated or would they have some sort of possible tag?

[/ QUOTE ]

I fit into this category as well. Only a fraction of percent from 67% WSD. I've been experimenting with rating these players with an Eagle (Tight with fair to good WSD stats), as well as a Mouse icon to denote players who have reasonably good VPIP, however they have terrible WSD stats. These (mouse) players seem generally good preflop, but undisciplined postflop.

I probably don’t have enough hands to really make any kind of significant observation, but the highest earners in terms of BB/100 fall within the Slightly Loose with Good WSD stats (averages 5bb/100 for “good” players vs. 7bb/100 for the slightly loose good players). I suspect many of these slightly loose but good players trade higher variance for a higher “potential” win rate. Not sure I can attach any significance to this just yet. I have 25,000 hands for myself and probably another few thousand “observed”.

[ QUOTE ]
I think your VPIP is really high for a good player

[/ QUOTE ]

This weekend I’ll look at dropping this to 22%. When I was experimenting and comparing stats earlier last week, I didn’t see much win/loss change for the higher 20-25% VPIP players so thought to include them as “good” but I certainly see your point. As I said, all this is preliminary and I’m completely open to suggestions.