LinearQLearningOptions<T>

Options & Configuration · AiDotNet.Models.Options

Configuration options for Linear Q-Learning agents.

Inheritance: ModelOptions → ReinforcementLearningOptions<T> → LinearQLearningOptions<T>

For Beginners

Linear Q-Learning extends tabular Q-learning to handle larger state spaces by using feature representations. Think of it as learning a formula instead of memorizing every single state.

Best for:

  • Medium-sized continuous state spaces
  • Problems where states can be represented as feature vectors
  • Faster learning than tabular methods
  • Generalization across similar states

Not suitable for:

  • Very small discrete states (use tabular instead)
  • Highly non-linear relationships (use neural networks)
  • Continuous action spaces (use actor-critic)

How It Works

Linear Q-Learning uses linear function approximation to estimate Q-values. Instead of maintaining a table, it learns weight vectors for each action and computes Q(s,a) = w_a^T * φ(s) where φ(s) are state features.

Constructors

Constructor Summary
LinearQLearningOptions()

Properties

Property Summary
int ActionSize { get; set; } Size of the action space (number of possible actions).
int FeatureSize { get; set; } Number of features in the state representation.