Opponent-Adaptive Bots
Learning objectiveUse observed tendencies with correct opportunities and sample sizes.
Understand the idea
VPIP is voluntary preflop investment divided by hands with an opportunity; PFR is preflop raises divided by eligible hands. Three-bet frequency uses actual re-raise opportunities. Aggression factor compares bets plus raises with calls. Fold-to-bet tendencies need the number of times a player faced that bet. Track observations by opponent identity, table, and season where appropriate.
A few hands are weak evidence. Shrink small samples toward a neutral baseline and require a minimum opportunity count before changing strategy. Separate public observations from hidden cards seen only at legitimate showdown. Never infer one opponent's private hand merely from a fold.
At the poker table
An opponent raised 2 of 3 observed opportunities. That 67% rate is unstable; after 200 opportunities it would be much more informative.
Code and decision sketch
Pseudocoderate = (observed_events + prior_events) / (opportunities + prior_weight)
# Keep numerator, denominator, and prior assumptions together.Common mistake
Do not divide three-bets by total hands or combine two different bots' histories under one name without a stable identity.
Try it yourself
Record five hands and identify which actually offered an opponent a three-bet opportunity.