Rating Other Players

[MyssGuy]

At approximately how many hands do you decide you can start rating other players? Can you start rating a person's play at 100 hands? 1k hands? I've got all PT and it's been great to help me evaluate my play, but how quickly can I start rating others? And how often should I update their rate?

[Avatar]

I see no reason not to start at 50. By then you have adequate enough hands to know if they are loose or tight and are starting to learn a bit about their pfr standards and aggression factors.

[Derek in NYC]

My cutoff is 30 hands. You should hit this in half an hour. If you wait too long, your ratings may be wasted on players no longer at your table.

Incidentally, I'm no statistician, but when I used to work with market research gurus, the rule of thumb those guys followed was that you could begin to make directionally correct characterizations of a population if the sample size was >30.

Some of your variables like VP$IP, CCPF, etc. will be pretty good at 30; others like AF, W$SD, etc. wont mean much.

But you will still have a good sense for whether somebody is a rock or a maniac after this many hands.

[MyssGuy]

I agree that a only small sample size is need to get an idea of play based PFR% and VP$IP, but 30 seems fairly small. I guess I'm used to thinking about my own play and needing tens of thousands to get an accurate rating....

[Munga30]

They need to be random samples from the population, not consecutive data points.

That being said, I use 30 as my cutoff, too, and it works fine. Better to see it earlier and know its fewer than not see it at all.

[Derek in NYC]

Why are these not random selections from the population? The randomness comes in the card randomization.

The only sample bias I see from taking consecutive hand histories is if the player in question is playing uncharacteristically that day, e.g., on tilt, feeling frisky, etc.

But even still, this bias is fine for short-run decisions, since you're only using the data to track the guy in the Gametime window.

[Chairman Wood]

I use 50 but I think as stated above 30 should be sufficent. When you talk about tens of thousands of hands to get good stats that applies to things like win rate which take a loooooooooooong time to converge. Numbers like PFR and VPIP converge much quicker and can be reliable enough for getting a read of something general like loose/tight after only 30-50 hands.

[SamIAm]

Why wouldn't you rate at 10 hands? I include the number_of_hands in my GameTime window, so I know not to count on the rating with a low sample-size, but what's the harm in getting the rough estimate?

I know that a low sample-size gives a lot more variance to your estimate, but it doesn't mess with the expected value at all. If I see somebody's played 9 of their first 10 hands, I'm not going to learn later that they only play 10 out of 100.

You're a thinking person. There's no reason to DENY yourself information, just because it COULD be innaccurate. As long as you don't fool yourself into blindly trusting the rating, extra information isn't bad.
-Sam

[housenuts]

does anyone have a good set of PT auto rate rules that they could export? that would be very helpful.