How many hands do I need for Poker Copilot’s HUD to be reliable

This is question you MUST be asking as a Poker Copilot user: can I trust the HUD data for a specific player? The answer is, it depends on how many hands you have AND which statistic you are looking at.

Most statistics in Poker Copilot are calculated by dividing a numerator by a denominator. For example, Voluntarily Put $ in Pot (VPiP) is calculated as follows:

Voluntarily put $ in pot % = 
       times voluntarily put $ in pot * 100 
       / hands played

If you’ve played 20 to 25 hands or so, you can consider this to be statistically significant. This is why Poker Copilot’s HUD initially shows statistics for a new player in grey, only switching to full-colour after you’ve collected 25 hands against that player.

The calculation for Folded to Four-bet Preflop (F4B) is

Folded to four-bet preflop % = 
       times folded to four-bet preflop * 100 
       / number of opportunities to fold to
four-bet preflop

For this to be as statistically significant  as VPiP after 25 hands, you need to play enough hands for the denominator to be >= 25. That is, 25 opportunities to fold to a preflop 4bet. How many hands is that? At full-ring microstakes No-Limit Hold’em, probably around 4500 hands.

As such I consider a stat such as F4B to be far more useful for analysing your own game, rather than making a decision about a villain. Before you made a decision based on an opponent’s F4B stat you should definitely click on their HUD panel to bring up the pop-up panel which has full information. In dark grey you’ll see the numbers behind the stat. Make sure that the denominator is at least 20 or so before using the stat to make a critical decision.

Screen Shot 2012 05 07 at 1 23 42 PM

 

You can see in the screenshot above that after 1,100 hands, that all the denominators are large enough except for “Folded to four-bet preflop”, “Folded big blind to steal attempt”, and “Folded to continuation bet on turn”. Here is a good rule of thumb for full-ring No Limit Hold’em for the most important stats:

Stats reliable after 25 hands: Voluntarily put $ in pot (VPIP),  Preflop raised (PFR)

Stats reliable after 150 hands: Aggression (Agg), Three-bet preflop (3bet)

Stats reliable after 500 hands: Continuation bet on flop (CBET), Folded to three-bet preflop (F3B)

Stats reliable after 1000 hands: Continuation bet on turn (TCBET), Four-bet preflop (F4B)

With less than this, be cautious, double-check the denominators, and use your own judgement about the player based on the stats that already are statistically significant.

In 6-max and heads-up, these statistics will become statistically significant quicker.

 

Coming soon: PFR by position, Aggression by street

I went on a statistic-adding rampage today:

Screen Shot 2012 05 04 at 5 07 07 PM

But that’s not all. Two new toolbar buttons allow you to turn on/off:

* VPIP/PFR by position

* the numbers behind the statistics, where appropriate.

There’s still plenty to do before I can release an update with these changes, as well as a lot of testing. Additionally I need to make sure performance is still acceptable with 30 new database fields per player and game needed for these changes. If you have 100,000 full-ring hands in your database, that’s roughly 30 million new pieces of information to store in the database. 

Comments on this Blog

In the last couple of weeks I tried an experiment with blog comments. They would go to the discussion forum as well as displaying on the blog posts. However it seems that the provider of our discussion forum still has some quality-control work to do with this feature. I’ve now gone back to the old commenting approach.

Sorry to anyone who has been confused as to the mismanagement of blog comments lately. Things are now back to normal.