PrioritizedReplayConfig<T>
Options & Configuration · AiDotNet.Configuration
Configuration for prioritized experience replay (PER).
How It Works
For Beginners: PER samples experiences based on their TD-error (surprise). High-error experiences are sampled more often because they have more to teach.
Constructors
| Constructor | Summary |
|---|---|
PrioritizedReplayConfig() |
Creates a new instance with default values. |
Properties
| Property | Summary |
|---|---|
T Alpha { get; set; } |
Alpha parameter controlling prioritization strength (0 = uniform, 1 = full prioritization). |
T FinalBeta { get; set; } |
Final beta value (should reach 1.0 by end of training). |
T InitialBeta { get; set; } |
Initial beta for importance sampling correction. |
T PriorityEpsilon { get; set; } |
Small constant added to priorities to prevent zero probabilities. |
int BetaAnnealingSteps { get; set; } |
Steps over which to anneal beta from initial to final. |