Textual exploitation stats for short NL, comment & discuss

[PeteX]

Ok, referring to my earlier thread about the textual stats ("Fit-or-fold, a textual postflop stat, what do you think?"), I decided to create a complete set of textual "exploitation" stats for preflop and postflop play.

I will specifically target these stats for short-handed NL cash games and I will use my short NL datamining database to calibrate the stats using players who have >1000 hands in the database.

Calibration principle: Most of the stats have two levels, the higher one marked with + and the lower one just with the text. The higher one marks the 90% threshold and the lower one marks the 75% threshold.

Example:
XYZ+, XYZ -> If XYZ+ is showing next to particular player, that stat is higher/lower (depending on the stat) than 90% of the other players and if just XYZ is showing, that stat is higher/lower than 75% of the other players. In other words, if the stat is visible (if it is not applicable, it will be not shown at all), it will tell you that regarding that stat, the player belongs to the top/bottom 10%/25% of the players when sorted by that stat.

I'll start with preflop stats and I would like to get some comments from short NL players as to what stats are relevant preflop in your opinion.

Here is a list of stats that I currently feel to be useful as preflop information. The grouping a...i means that the abbreviations in the group are mutually exclusive, you see either one of them or none at all. You would then create a HUD group (or perhaps 2 groups) with these stats and you would see something between none of them and all of them, depending on how exploitable the player is. So, if someone is REALLY weird fish, you could see up to 9 (one from each group a...i) different abbreviations next to his seat on the hud. If someone is ABC-TAGish, you might see none of the abbreviations. In which case it might be time to change tables...

a) F3B, F3Bs, F3Bns
Folds to 3 bet when open-raised (F3B = always, F3Bs = stealing, F3Bns = non-stealing)

b) FRL+, FRL
Folds to raise when limped

c) NCC+, NCC
Never cold-calls (=belongs to bottom 10%/25% cold-callers)

d) 3Bet+, 3Bet, N3Bet+, N3Bet
3Bets often / Never 3bets (if you see "N3Bet+" and the player 3-bets, HE HAS A HAND!)

e) NBD+, NB, DBD, BDc, BDr
No blind defense / Defends big blind (specifically with call / raise)

f) RL+, RL
Raises limper(s) very light (+) or light

g) SQ+, SQ
Squeezes light

h) FSQ+, FSQ
Folds to a squeeze very easily or easily

i) ST+, ST, NST+, NST
Steals often / Never steals

For example, taking the stat a) as an example:
- If the "fold to 3 bet when open raised from any position"-statistic places the player among the top 25% players who fold the most, then you would see "F3B" next to this player.
- If the "fold to 3 bet when open raised from a stealing position"-statistic places the player among the top 25% players who fold the most, then you would see "F3Bs" next to this player.
- If the "fold to 3 bet when open raised from a non-stealing position"-statistic places the player among the top 25% players who fold the most, then you would see "F3Bns" next to this player.

So, if you see F3B, F3Bs or F3Bns next to the player and he has open-raised, you might consider 3-betting light depending whether he was stealing or not. If you don't see any of these figures next to the player, then he is not going to fold that often on a 3-bet.

NOTE: When using these stats, there are NO NUMBERS to look at, these are just abbreviations that are meant to convey you the information regarding the exploitability of a certain player. The absence of an abbreviation means that the player is not exploitable by that particular stat. Obviously you can also have numbers on the Hud, but my point here is to try and get rid of MOST of the numbers and try to pinpoint possible exploitation opportunities with simple and easy to read textual abbreviations. I'd guess that in addition to these abbreviations (plus the postflop stuff that I'll get into later), I'd probably use only VPIP, PFR and AFq to form a general opinion of the player.

So these would be the textual values that you would see on the hud. I'll be using color ranges to indicate the reliability of the stat like this:
Red : < 10 samples of the situation seen
Yellow: 10...30 samples of the situation seen
Blue: 30...50 samples of the situation seen
Green: >50 samples of the situation seen

So, if you see Green "NST+" on a player who is stealing your blind, you can be pretty sure that he really has a hand. With green we have seen the player having >50 steal opportunities but he has tried to steal only as little as the bottom 10% of the stealers on my short NL datamining database. With green NST, he would be in the bottom 25% of the stealers.

My intention is to collect all these stats into a single zip-file and perhaps even create a sample layout for them and then donate those to the repository.

But before that, I'd like some comment from short NL regulars about my selection of stats. Are all of those useful? Would you add/change/delete something?

[bloodndef]

I love this idea.

Textual stats have been on my mind lately as well.

Presently I am creating a HUD popup that displays hand ranges taken from known hands.
For example if a player has deep-limped with AA and this hand is known (i.e. went to showdown or was voluntarily shown) I have a grid of stats that displays all the known hands that were LWPC (limped with previous callers). If the player has hands that went to showdown after LWPC you will be able to see a percentage stat that is the result of this:

(cnt_p_lwpc_premipr / cnt_p_lwpc_known) * 100

where cnt_p_lwpc_premipr is a count of how often a premium pair (AA-KK) is shown down after limping with previous callers preflop and cnt_p_lwpc_known is a count of all hands that are known after limping with previous callers preflop.

I've divided my HUD popup into 7 columns that depict 7 hand ranges: Premium Pairs, Solid Pairs, Small Pairs, Premium Unpaired, Solid Unpaired, Suited Connectors, and Light Hands. My rows are preflop actions like LWPC, LFI, RFI, etc. (some are also separated into positional stats)

Is this the best way to do this I'm wondering, or would there be a textual way that is more clearly identifiable in a HUD popup? (E.g. Every time a player has limped first in (LFI) in early position with AA, a textual AA appears in an appropriate HUD popup row that would say LFI-EP: AA.) I fear that with a large sample size, this method could get cluttered, and the simple percentage display might be easiest and most effective. Is there another way to represent this textually?

Can you post an example of how you put one of these together?

Keep it up!

[kraada]

These stats sound really great, and I look forward to seeing them.

I did much some simple textual stats in the How To: Advanced HUD Configuration if you're looking for simple textual stats examples.

[PeteX]

Can you post an example of how you put one of these together?

Sure, it's pretty simple!

Create a new custom stat like you would create any other stat, but put just 0 to the expression.

And then create the actual stat formula using the if(condition,trueresult,falseresult)-statement(s) in the Format Expression-field.

For example, I have here "Limp/Fold Preflop"-stat and this is what I have on the "Format Expression"-field:

Code: Select all

if (cnt_limp_fold / cnt_limp_face_raise > 0.8, 'FRL+',
if (cnt_limp_fold / cnt_limp_face_raise > 0.6, 'FRL', ''))

Nothing magical here:
- If the value of "cnt_limp_fold / cnt_limp_face_raise" exceeds 0.8, then "FRL+" is displayed on the HUD.
- If the value of "cnt_limp_fold / cnt_limp_face_raise" exceeds 0.6, then "FRL" is displayed on the HUD.
- Otherwise nothing is displayed on the Hud for this stat.

So, basically you are generating the textual value on the Hud using the FORMATTING of the stat, NOT the stat value itself. Note that those values there are not calibrated yet properly, but you get the idea.

Obviously the nested structure of the IF-statements can get quite cluttered, but that's the price you pay for this technique and the "Validate"-link works just fine to test if you have properly nested the if-statements. After that it's just up to you to come up with proper conditions to display various abbreviations on the hud. I prefer to make such abbreviations which REALLY make the player stand out from the crowd in some respect, that's why I have the 10%/25% thresholds.

[bloodndef]

Thanks PeteX and Kraada!

Here is a small screenshot of a stat I made called LFI-HJPL (limped first in, from the Hijack or Plank, positions 2 or 3) across 7 different hand ranges.


I have the actual ranges appear textually color coding the frequency from red to green, lowest to highest.
I am in the process of making a custom HUD popup (which I'm sure will look better with a black background).

Textual stats are certainly helpful and I'm very excited to see your textual exploitation stats, PeteX.

Does anyone know how to make the background of my stats black in the reports section?

Thanks again!

[kraada]

That looks pretty awesome!

I'm curious: what ranges are you using for those various stats (ie- which hands count as what)? I tried to come up with groups that I was happy with, but I kept wanting to change them for one reason or another and just never got happy with any specific definition.

As far as I know there isn't a way at the moment to change the background in a PT3 report from white but I'll look into ways around it for you.

[WhiteRider]

PT3 uses Windows colours for reports so you would need to change your settings in Windows.
I believe there are plans to allow configurable colours, though.

[PeteX]

A short note on the progression of these stats (to anyone that might be interested):

I decided to change my criteria for players to use in the analysis from having >1000 hands in the database to such that I'll use players that have seen a particular situation >10 times. For example, if a player has tried to steal blinds and faced a 3-bet from sb/bb at least >10 times, then he will be included in the analysis.

However, I learnt that my datamined NL short database was way too small, I only got around 30 players (I want to have at least 100 players for each stat) for the above example criteria, so I'm now building a much bigger database, should be ~30mil hands. The thing is just that my current hd is so full of stuff that I can't fit that database into it so today I'm going to buy a new 1,5T hd that will be dedicated just for PostgreSQL databases and after I manage to build by NL short analysis database, I'll start calibrating the stats. Importing the hands will take quite a while, last PT3 estimate was ~98 hours (I let the import run for couple of days on my current hd but then noticed that I will not have enough space eventually so decided to abort mission there) for the ~80thou hand history files that I have for short NL.

I have similar databases for full-ring limit and short limit, so I might do similar stats (vary them a bit, e.g. noboby will fold to a 3bet on short limit so no use to have such stat for preflop) for them later as well.

[kraada]

Thanks for the update. I think it's worth noting generally how many hands you need to have a remotely accurate sample!

[PeteX]

Just formatting my new Seagate Barracuda 7200.11 1,5T HD...

But oh man the formatting takes ages...