The Limits of Datamining

[jason1990]

Here's a hypothetical situation:

Suppose I am an expert online limit holdem player. I play at a certain exclusive site. There are no new players on this site, only regulars. I have 1000 hands in my PT database on each player.

Suddenly, someone gives me a new database with 10000 hands on each player. Will I be able to use this data to significantly increase my win rate?

[AaronBrown]

It depends on the players.

You can figure out all you need to know about predictable players from the first 1,000 hands. You can't figure out really good players with any amount of data. In either case, the extra data won't help.

In between, you could use the extra information to detect patterns. For example, suppose a guy raises preflop 5% of the time. With 1,000 hands, you'll see 20 examples, maybe three or four of these will go to showdown. That's clearly not enough to get the full rangs of hands he raises preflop on, and doesn't even scratch the surface of figuring out how he changes that based on position, previous action and other factors. It won't tell you whether he's consistent, erratic or deceptive.

For most cases, the extra information doesn't help much. Mainly you want to know if a guy raises preflop a lot or a little, how often he tends to bluff, and whether he slowplays his strongest hands. You can figure that out from 1,000 hands if you can figure it out at all. But knowing that he tends to raise with AKs but not AKo could make or save you a bundle once every few thousand hands. Knowing a lot of stuff like this could make or save you a bundle every hundred hands or so. These things would take a lot of hands to figure out.