class Crowbar::Evolution::Bandit

Overview

Multi-Armed Bandit credit assignment using the Upper Confidence Bound (UCB1) algorithm. Dynamically balances exploitation of high-yield mutators/scopes with exploration of less-tested ones.

Defined in:

crowbar/evolution/bandit.cr

Constructors

.new(exploration_coeff : Float64 = 1.41421356)

Instance Method Summary

#arms : Hash(String, Arm)
#ensure_arm(name : String) : Arm
Ensure an arm is registered
#exploration_coeff : Float64
#exploration_coeff=(exploration_coeff : Float64)
#pulls_for(name : String) : Int64
#record_pull(name : String)
Record that an arm was executed
#reward(name : String, value : Float64)
Credit an arm with a reward
#rewards_for(name : String) : Float64
#score_for(name : String) : Float64
Retrieve score for inspection/telemetry
#select(available_arms : Array(String), prng : PRNG) : String
Select the best arm among available names based on UCB1 score
#total_pulls : Int64
#total_pulls=(total_pulls : Int64)

Constructor Detail

def self.new(exploration_coeff : Float64 = 1.41421356) #

Instance Method Detail

def arms : Hash(String, Arm) #

def ensure_arm(name : String) : Arm #

Ensure an arm is registered

def exploration_coeff : Float64 #

def exploration_coeff=(exploration_coeff : Float64) #

def pulls_for(name : String) : Int64 #

def record_pull(name : String) #

Record that an arm was executed

def reward(name : String, value : Float64) #

Credit an arm with a reward

def rewards_for(name : String) : Float64 #

def score_for(name : String) : Float64 #

Retrieve score for inspection/telemetry

def select(available_arms : Array(String), prng : PRNG) : String #

Select the best arm among available names based on UCB1 score

def total_pulls : Int64 #

def total_pulls=(total_pulls : Int64) #