Working on a 3-bet light stat

[bloodndef]

I'm working on a stat that shows how often a player has 3-bet light preflop (hands NOT: AA-88, AKs-ATs, AKo-AQo, and KQs).

(cnt_p_3bet_light / cnt_wtsd_after_p_3bet) * 100

Anyone done something similar? Point me in any good directions?

[bloodndef]

cnt_p_3bet_light


sum(if[holdem_hand_player_statistics.flg_p_3bet AND NOT holdem_hand_player_statistics.id_holecard=1 AND NOT holdem_hand_player_statistics.id_holecard=2 AND NOT holdem_hand_player_statistics.id_holecard=3 AND NOT holdem_hand_player_statistics.id_holecard=4 AND NOT holdem_hand_player_statistics.id_holecard=5 AND NOT holdem_hand_player_statistics.id_holecard=6 AND NOT holdem_hand_player_statistics.id_holecard=8 AND NOT holdem_hand_player_statistics.id_holecard=26 AND NOT holdem_hand_player_statistics.id_holecard=27 AND NOT holdem_hand_player_statistics.id_holecard=49 AND NOT holdem_hand_player_statistics.id_holecard=70 AND NOT holdem_hand_player_statistics.id_holecard=89 AND NOT holdem_hand_player_statistics.id_holecard=106 AND NOT holdem_hand_player_statistics.id_holecard=121, 1, 0])

[bloodndef]

cnt_wtsd_after_p_3bet

sum(if[holdem_hand_player_statistics.flg_showdown AND holdem_hand_player_statistics.flg_p_3bet, 1, 0])

[bloodndef]

Fixed it.

cnt_p_3bet_light

should be:

sum(if[holdem_hand_player_statistics.flg_p_3bet AND holdem_hand_player_statistics.id_holecard > 0 AND NOT holdem_hand_player_statistics.id_holecard=1 AND NOT holdem_hand_player_statistics.id_holecard=2 AND NOT holdem_hand_player_statistics.id_holecard=3 AND NOT holdem_hand_player_statistics.id_holecard=4 AND NOT holdem_hand_player_statistics.id_holecard=5 AND NOT holdem_hand_player_statistics.id_holecard=6 AND NOT holdem_hand_player_statistics.id_holecard=8 AND NOT holdem_hand_player_statistics.id_holecard=26 AND NOT holdem_hand_player_statistics.id_holecard=27 AND NOT holdem_hand_player_statistics.id_holecard=49 AND NOT holdem_hand_player_statistics.id_holecard=70 AND NOT holdem_hand_player_statistics.id_holecard=89 AND NOT holdem_hand_player_statistics.id_holecard=106 AND NOT holdem_hand_player_statistics.id_holecard=121, 1, 0])

[bloodndef]

This stat is for everyone but hero.

Hero would be:

(cnt_p_3bet_light / cnt_p_3bet) * 100

[bloodndef]

I feel like we're on the verge of some amazing stuff with the hand-range-type stats. I've got a player with 1.2k hands whose 3bet pf stat is 4.98 and whose 3-bet light stat is 66.67 - showing that he is not 3-betting tightly but rather SPORADICALLY. Hands like K7o and 98o are in his range even though his 3bet stat is a relatively solid 4.98. Another player with a similar 3bet pf stat shows 0/4 in the 3bet light stat, showing that he is most likely 3betting tightly.

I can't wait to break this 3 bet light stat down by positions. Could also do a PFR light stat. And also have stats with varying degrees of "lightness" i.e. a PFR with trash hands stat from 3 off the button.

[WhiteRider]

That looks really interesting - good work!

One suggestion - cnt_p_3bet will include all hands, not just those with known hole cards, so the stats could be a bit skewed.
You might want to create a new column to count 3-bets with known hole cards.
If you don't do this you could get vastly different results for players who raise light a lot but then fold easily after that.
In fact, saying that, I think you'll always struggle to get this very accurate because when people are 3-betting light they are much less likely to get to show down so you won't see the cards very often.

When you're happy with what you've got, if you want to share your stats you can submit them to the Repository.

Trivial point - you could shorten your expression slightly since you are checking hole cards 1 to 6 individually, you could use > 0, < 7.

[bloodndef]

Thanks!

How can I look up hands with known hole cards? Is my column cnt_wtsd_after_p_3bet effectively doing the same thing? This was my goal.


I appreciate all your feedback!

And I certainly want to make some contributions to the repository in the next week or so, if you guys see stuff you like.

[WhiteRider]

I think that's what your check for holcards > 0 is doing, but of course that isn't included in the built in cnt_p_3bet column.
You could build a duplicate of that and add your check for hole cards > 0.

However, I think the problem is going to be that you might have someone who 3-bets really light (so cnt_p_3bet is really high for the number of opportunities to 3-bet) but if they fold to resistance if they miss, you're likely to only see the stronger hands going to showdown most of the time - so you might find that your 3bet light stat actually gives a really low figure.
Granted, as your hand sample goes up you will see some hands where they got lucky with trash and went to showdown, but I think it will take a long time for this to become representative.

As an example, let's say you have someone who 3-bets every time he gets the chance.
His "3Bet Preflop" stat will be 100%.
Take a sample of hands, though.
1. He 3 bets and everyone folds. No 3-bet light because we don't see his hands.
2. His opponent reraises and he folds - presumably he was 3-betting light, but again we don't see his cards so it won't count in your stat.
3. His opponent reraises and he calls with AK. This isn't a 3-bet light but will count in the denominator of your stat.
4. His opponent calls but then folds to a c-bet on the flop. Again, no cards = no data for stat.

Out of these hands, the only one that you have his card information on is #3, and he isn't 3-betting light there, so his 3-bet light stat will be 0% (0/1).
Since we don't know his cards he could have had AA or trash for 1 and 4, although he's probably light in 2 since he folds.
We could add several more of these examples, and we probably wouldn't get a showdown with a trash hand for quite a while.
If you count all the 3-bets in your denominator, then you have 0/4 for light raises. If you only count hands with known cards, you get 0/1.
However the fact remains that he 3-bet 4/4, which implies he's light overall (especially if I could be bothered to type a few more examples...).

I'm not being deliberately awkward, but I think these kind of stats are always going to be skewed so you need to be careful how you interpret the values.

You might be able to do something with how often they go to showdown after 3-betting as well.

[bloodndef]

Thank you sir, you are spot on.

I know this stat needs a very large sample size and is most effective when combined with the normal 3bet stat. But this was just the beginning. I was trying to see if I could even make such a stat with my limited knowledge. Other stats such as limping ranges, CCPF ranges, PFR ranges, etc. all combined with positional stats are all possible for me now, and I am beginning to put them together.

Thanks again.