Question/Problem in Advanced HUD Configuration

[preparac]

Hi Krada

I’m trying to adapt the stats you present in your Advanced HUD Configuration-thread to my tourney custom stats and looking closely at what you have done (thanks for your great guide by the way!!!) I encountered some problems/questions with the 3betting stats, which I want to report you:

(1)
a) Your column cnt_p_3bet_nvs_bs counts
sum(if[not(holdem_hand_player_statistics.flg_blind_def_opp) and holdem_hand_player_statistics.flg_p_3bet, 1, 0])
is not exactly what it ment to be (Number of times player 3 bet preflop when not against a blind steal), because it counts also 3bets from the BU against open raises from the CO and therefore should be described as: Preflop 3Bet Percentage when not in the blinds OR when in the blinds NOT vs a blind steal attempt.

b) Therefore:
3bet Preflop NvBS – colored” + 3bet vs steal % > 3bet Preflop, because this BU 3bets CO cases are counted double. Am I right here? At least this is the result I get in my reports, using your stats...

(2)
a) Your stat 3bet vs steal % uses a misleading value expression I think
((cnt_bu_3bet_steal / cnt_bu_3bet_steal_opp) * 100 + (cnt_bb_steal_raise / cnt_steal_def_opp_bb) * 100 + (cnt_sb_steal_raise / cnt_steal_def_opp_sb) * 100) / 3

This expression doesn’t weight the number of opportinitys in each position and leads therefore to some strange results (imagine a player 3bet his only 3bet opportunity in the BU against a CO-Steal but folded 10 of 10 3Bet opportunitys in the SB and the BB each against Steal attempts, this expression will show the value 33%)!

b) So a better way to account for the different number of 3Bet opportunities in the different positions would be the valu expression
((cnt_bu_3bet_steal + cnt_sb_steal_raise + cnt_bb_steal_raise) / (cnt_steal_def_opp_bb + cnt_steal_def_opp_sb +cnt_bu_3bet_steal_opp) * 100

c) In the light of the before mentioned problem (1) AND because for my taste the Steal Attempt definition is a little bit arbitrary for this 3bet cases (why not to count also an open raise from the HJ as a steal attempt?) AND because we anyway have much less resteal opportunitis in the Button compared with the SB and the BB, I prefere to change your definition of 3bet vs steal % a little bit by restricting it on the cases where I’m sitting in the blinds:

3bet PF vs St (Preflop 3Bet Percentage vs a Blind Steal Attempt when in the Blinds)
((cnt_bb_steal_raise + cnt_sb_steal_raise )/ (cnt_steal_def_opp_bb + cnt_steal_def_opp_sb) * 100)

(3)
Coloring is really a nice thing, but in this case I don’t get what info you exactly win when you color the 3betting frequency not against steals with the raise frequency against steals …? I prefer to see the exact 3bet-stats for both cases (against steals and not steals) and then to color them with the reaction of the player to a (re-resteal) 4bet.

Therefore I build for the coloring the stats
Fold to PF 4Bet after 3Bet nvs St and (Fold to Preflop 4Bet Percentage after 3Betting NOT against a Blind Steal Attempt)
Fold to PF 4Bet after 3Bet vs St (Fold to Preflop 4Bet Percentage after 3Betting against a Blind Steal Attempt)
Respectively

With the value expression (for the first case)
(cnt_p_4bet_def_action_fold_after3bet_nvs_bst / cnt_p_4bet_def_opp_after3bet_nvs_bst) * 100

When
cnt_p_4bet_def_action_fold_after3bet_nvs_bst
has the expression
sum(if[tourney_holdem_hand_player_statistics.flg_blind_def_opp = false AND tourney_holdem_hand_player_statistics.flg_p_3bet AND tourney_holdem_hand_player_statistics.enum_p_4bet_action='F', 1, 0])

and

cnt_p_4bet_def_opp_after3bet_nvs_bst
has the expression
sum(if[tourney_holdem_hand_player_statistics.flg_blind_def_opp = false AND tourney_holdem_hand_player_statistics.flg_p_3bet AND tourney_holdem_hand_player_statistics.flg_p_4bet_def_opp, 1, 0])

I admit that like this you need quite a lot of hands until you get significant coloring stats, but besides of this downside the info you get from this combination (3betting light against steals then folding often to 4bets) seems to me quite straight forward...

thanks for feedback :)

[kraada]

Hello :)

(1)
a) Your column cnt_p_3bet_nvs_bs counts 3bets from the BU against open raises from the CO and therefore should be described as: Preflop 3Bet Percentage when not in the blinds OR when in the blinds NOT vs a blind steal attempt.

This is true. There isn't an easy way to filter out those spots from the definition so I left it in. Since I also have 3bet in late position you should be able to get an idea if the guy is raising a ton on the button (as 3bet in late position will also pick up all of these spots), but if you think about the sample size the times he has a chance to 3bet from the button vs a cutoff steal are going to be so much fewer than the times that he's going to have a chance to 3bet otherwise not vs a blind steal that I don't think it taints the sample much.

Consider a player you have 9999 hands on. Assuming he played all full orbits, that's 1111 times on the button. Just to make the math easier, assume all villains open up 5% in EP, 10% in MP and 15% in LP. There are 2 EP spots so 111 times it will be opened in EP and this won't matter. Of the remaining 1000 times, 200 will be opened in MP. Of the remaining 800 times, 120 will be opened by the HJ. So there are only 680 opportunities for CO here. If CO opens a huge range, say, 40% that's still only 272 3bet opportunities from the button. He has 111+200+120 or 431 3bet opportunities from other positions.

In the SB using the same math he'll face 431 3bet spots, same for the BB and the CO. In the HJ he has 311 3-bet opportunities. In MP2 he has 211, MP1 111 UTG+1 55. So he'll have 431*3 + 311 + 211 + 111 + 55 = 1981 3-bet spots that are clean. Plus the 431 from other spots is 2412 3-bet opportunities. And there are 272 times he can 3-bet vs a CO open steal.

Let's give this guy a very polarized 3bet range. He'll 3bet the top 3% of his hands in all spots unless he's on the button vs a CO steal, then he 3bets the top 10% of his hands. What would his total 3bet PF NvBS look like?

.03 * 2412 = 72, .10 * 272 = 27. So he'll have 3bet 99 times out of 2984 opportunities, or 3.3%.

I realize that was long winded but the point is that you face so many 3bets in other spots that it doesn't make a large change here.

(as a side note: 3bet in LP for this player would be 5.5%, so you would notice that there's a definite jump in late position 3betting. I'll leave the math to you.)

3bet Preflop NvBS – colored” + 3bet vs steal % > 3bet Preflop, because this BU 3bets CO cases are counted double. Am I right here?

Yes, but I hope I've explained why that doesn't bother me all that much :)

((cnt_bu_3bet_steal / cnt_bu_3bet_steal_opp) * 100 + (cnt_bb_steal_raise / cnt_steal_def_opp_bb) * 100 + (cnt_sb_steal_raise / cnt_steal_def_opp_sb) * 100) / 3

I'm not sure where you got this stat, but I don't believe it's mine. I use Raise vs Steal Att, which is calculated like you say, except for button 3bets vs CO steals. If there were an easy way to pick out button defenses against a CO steal I'd add that in, but there isn't an easy way to pick out those spots, so they've been ignored.

c) In the light of the before mentioned problem (1) AND because for my taste the Steal Attempt definition is a little bit arbitrary for this 3bet cases (why not to count also an open raise from the HJ as a steal attempt?) AND because we anyway have much less resteal opportunitis in the Button compared with the SB and the BB, I prefere to change your definition of 3bet vs steal % a little bit by restricting it on the cases where I’m sitting in the blinds:

3bet PF vs St (Preflop 3Bet Percentage vs a Blind Steal Attempt when in the Blinds)
((cnt_bb_steal_raise + cnt_sb_steal_raise )/ (cnt_steal_def_opp_bb + cnt_steal_def_opp_sb) * 100)

That's exactly the same as my Raise vs Steal Attempt. I'm glad we're on the same page :)

(3)
Coloring is really a nice thing, but in this case I don’t get what info you exactly win when you color the 3betting frequency not against steals with the raise frequency against steals …? I prefer to see the exact 3bet-stats for both cases (against steals and not steals) and then to color them with the reaction of the player to a (re-resteal) 4bet.

What I get out of it is a way to tell quickly if the person is 3betting lighter vs steal attempts. I've been considering changing the scale on that coloring to every 2% instead of every 10% -- if someone is 3betting vs steals more than 10% it's a huge range and I should just know to 4bet for value anything realistic and just never steal with trash. Anything significantly different from the blinds gives me less incentive to steal against them as they're more likely to fight back (unless they just fold the rest of the time, then I can still steal with some frequency).
statistics.flg_p_4bet_def_opp, 1, 0])

the info you get from this combination (3betting light against steals then folding often to 4bets) seems to me quite straight forward...

That's the great thing about the PT3 HUD you can customize it so much -- you want to see the times someone 3bets light vs steals and then folds to a 4bet, I want to just see the times they 3bet light. Each has value, I'm not going to say that yours has no value and I think you'll agree that knowing how likely someone is to 3bet light is useful. I agree that the sample you'd need before folding to 4bets vs a resteal from the btn/co would converge on anything close to reality would be huge. But you agree on that also. So long as you use the stats with that in mind, it's not going to cause any problems. You can imagine though that someone who didn't understand that could end up losing a stack because of it, which is why I didn't include it in the default.

I think I'll look into adding a "fold to 4bet having previously 3bet" stat by position into my popup though. You've convinced me that it'd be awfully useful in specific spots, and stats that are awfully useful in specific spots are exactly those I want in a popup window.

[preparac]

... but if you think about the sample size the times he has a chance to 3bet from the button vs a cutoff steal are going to be so much fewer than the times that he's going to have a chance to 3bet otherwise not vs a blind steal that I don't think it taints the sample much.

This is absolutely correct. I checked a few players in my database and found also that only a few % of all 3bets against steal-raises are BU-3bets vs CO-steals ...

I'm not sure where you got this stat, but I don't believe it's mine.

...Strange ... I thought I downloaded this stat from the zip-file you attached to your "advanced"-guide ... But if you tell it does'nt come from there then I probably downloaded it months ago from the repository and I don't remeber anymore...

What I get out of it is a way to tell quickly if the person is 3betting lighter vs steal attempts. I've been considering changing the scale on that coloring to every 2% instead of every 10% -- if someone is 3betting vs steals more than 10% it's a huge range and I should just know to 4bet for value anything realistic and just never steal with trash. Anything significantly different from the blinds gives me less incentive to steal against them as they're more likely to fight back (unless they just fold the rest of the time, then I can still steal with some frequency).

Yeah, I got your point, but for the exact reason that already a difference of a few % MAKES a hughe difference, I prefere to see the the two numbers actualy side by side. But then again, I have players in my database, who are 3betting far to light against steal attemps (> 12% over several thousand of hands) but still NEVER fold to a 4bet, so against THESE players I really want to play fast my top 5% PF and play cautiously with the rest...

I agree that the sample you'd need before folding to 4bets vs a resteal from the btn/co would converge on anything close to reality would be huge. But you agree on that also.

Yea, I guess you need at least 3-4000 hands of somebody who 3bets light ... and even more hands of a tighter player ... that's the downside

Thanks again for your insights.

May I ask you here another question about (positional stats in) your advanced HUD guide instead of opening a new thread:

You made a rather complex definition schema of EP, MP and LP, where this definition changes with the number of players at the table. Where you see the advantage versus the conventional definition of LP as postions 0/1/2, MP 3/4 and EP 5/6/7 regardles of the number of players at the table (i.e. at a short handed table you don't have EP at all ...)? For my taste the simple straight forward definition copes better wiht the caracter of the game than your relative definition, because in a short handed game you should play UTG like MP or even like LP ... Or do I miss something?

[kraada]

Yeah, I got your point, but for the exact reason that already a difference of a few % MAKES a hughe difference, I prefere to see the the two numbers actualy side by side. But then again, I have players in my database, who are 3betting far to light against steal attemps (> 12% over several thousand of hands) but still NEVER fold to a 4bet, so against THESE players I really want to play fast my top 5% PF and play cautiously with the rest...

I agree that a few % makes a huge difference and that's why I was thinking of changing the default color range from 0-10 for red to 0-1 red, 1-2 light red, 2-3 orange, etc. I just haven't had a chance to play with it that much. I think in the long run that will be the way to go, though people just starting out with the HUD might be confused why cyan is 100% in some places and 10% here. (Though if they think about the stat much they should realize anybody 3-betting 10% vs a steal is 3-betting a huge range and likes burning money :)

May I ask you here another question about (positional stats in) your advanced HUD guide instead of opening a new thread:

You made a rather complex definition schema of EP, MP and LP, where this definition changes with the number of players at the table. Where you see the advantage versus the conventional definition of LP as postions 0/1/2, MP 3/4 and EP 5/6/7 regardles of the number of players at the table (i.e. at a short handed table you don't have EP at all ...)? For my taste the simple straight forward definition copes better wiht the caracter of the game than your relative definition, because in a short handed game you should play UTG like MP or even like LP ... Or do I miss something?

I'm always willing to answer questions :)

There are a few reasons why I did this. First, there are flags in the database for EP/MP/LP. They're not usable in stats at the moment, but they will be eventually. My definition is consistent with those flags, and I thought it would be good to be consistent that way later if you can just check in a simple fashion using lookup_positions whether you're in MP you don't have to re-learn what an MP range is at a 6 max table.

Second, I didn't want to have 9 columns. If I'm going to be doing it by position in an absolute fashion it seems like it'd be necessary to have a column for each position, or combine them every two or some such. The latter seems too arbitrary for my taste and the former is too cluttered. I don't think that many people differentiate between positions 6 and 7 at a full ring table, so lumping things together seems good to me. It also shortens the amount of hands you need to see before you can have a real read on someone.

Third, with regards to short handed tables, I still think it's beneficial to chop it up into more pieces. Ideally we'd love to have stats on every single spot we can in the easiest to read fashion. If people don't play UTG that differently from UTG+2 at 6-max that will show up in the split I have pretty quickly and you'll know this player isn't positionally aware. Though I admit that 6-max plays very differently from full ring. Personally I play full ring, so I can't say 100% for sure that this definition works fantatsically for 6-max, but I trust that when the database schema was built some serious thought went into the different definitions.

[preparac]

There are a few reasons why I did this. First, there are flags in the database for EP/MP/LP. They're not usable in stats at the moment, but they will be eventually. My definition is consistent with those flags, and I thought it would be good to be consistent that way later if you can just check in a simple fashion using lookup_positions whether you're in MP you don't have to re-learn what an MP range is at a 6 max table.

... good to know. Thanks for the info.