verve::defaults Namespace Reference

Contains all default values used to initialize variables. More...

Variables
const real	stepSize = (real)0.1
const real	eTraceTimeConstant = (real)0.1
const real	TDDiscountTimeConstant = (real)1.0
const real	valueFunctionLearningTimeConstant = (real)0.1
const real	policyLearningMultiplier = 5
const real	modelLearningTimeConstant = (real)0.001
const real	activeETraceThreshold = (real)0.01
const real	minActionSelectionProb = (real)0.0
const AgentArchitecture	agentArchitecture = RL
const unsigned int	maxNumPlanningSteps = 10
const real	planningUncertaintyThreshold = (real)0.1

---

Detailed Description

Contains all default values used to initialize variables.

---

Variable Documentation

const real verve::defaults::activeETraceThreshold = (real)0.01

TDConnections with eligibility traces below this threshold are considered inactive, and their eligibility traces get set to zero. Definition at line 135 of file Defines.h. Referenced by verve::ActiveTDConnectionList::decayETraces().

const AgentArchitecture verve::defaults::agentArchitecture = RL

The default Agent architecture to use. Definition at line 144 of file Defines.h. Referenced by verve::AgentDescriptor::AgentDescriptor().

const real verve::defaults::eTraceTimeConstant = (real)0.1

Determines how fast eligibility traces change. Definition at line 110 of file Defines.h.

const unsigned int verve::defaults::maxNumPlanningSteps = 10

The maximum number of steps to take during a planning sequence. Definition at line 147 of file Defines.h. Referenced by verve::AgentDescriptor::AgentDescriptor().

const real verve::defaults::minActionSelectionProb = (real)0.0

The minimum probability of choosing each action. This should be kept above zero to ensure that exploratory actions are never totally ignored. This value times the number of actions MUST be < 1. Definition at line 141 of file Defines.h.

const real verve::defaults::modelLearningTimeConstant = (real)0.001

Determines the learning rate for the predictive model. This is how long it takes (in seconds) for errors to be reduced to 37% of their initial values. Definition at line 130 of file Defines.h.

const real verve::defaults::planningUncertaintyThreshold = (real)0.1

The maximum amount of estimated uncertainty to tolerate before ending a planning sequence. Definition at line 151 of file Defines.h. Referenced by verve::AgentDescriptor::AgentDescriptor().

const real verve::defaults::policyLearningMultiplier = 5

The policy's learning rate is a combination of this multiplier and the value function's learning rate. Definition at line 125 of file Defines.h.

const real verve::defaults::stepSize = (real)0.1

Update step size. Definition at line 65 of file Defines.h.

const real verve::defaults::TDDiscountTimeConstant = (real)1.0

Determines how much future rewards are discounted. For example, a discount time constant of 1 means that rewards received 1 second in the future are worth only 37% of what they are worth right now. Definition at line 116 of file Defines.h.

const real verve::defaults::valueFunctionLearningTimeConstant = (real)0.1

Determines the learning rate for TDConnections in the value functions. This is how long it takes (in seconds) for errors to be reduced to 37% of their initial values. Definition at line 121 of file Defines.h.

---