Search Results for "stochastic"

...It includes dynamic programming methods for finite environments and tabular temporal-difference learning with SARSA and Q-learning. A Deep Q-Learning agent extends Q-learning to continuous state features through neural-network function approximation. The project also experiments with stochastic and deterministic policy gradients and actor-critic methods for continuous action spaces. Its API separates reusable neural-network utilities from reinforcement-learning agents. Interactive Gridworld, PuckWorld, and WaterWorld demos make the algorithms easier to inspect visually. The library is primarily educational and exposes the learning loop in compact JavaScript code.