Brett Daley's repositories
dqn-lambda
NeurIPS 2019: DQN(λ) = Deep Q-Network + λ-returns.
gym-classics
Classic environments for reinforcement learning and dynamic programming, implemented in OpenAI Gym and Gymnasium.
stratified-experience-replay
Stratified Experience Replay. Correcting Multiplicity Bias in Off-Policy Deep Reinforcement Learning. AAMAS 2021.
virtual-replay-cache
Virtual Replay Cache. A modified DQN(λ) implementation with a significantly reduced memory footprint.
averaging-nstep-returns
ICML 2024: Averaging n-step Returns Reduces Variance in Reinforcement Learning
expectigrad
A deep learning optimizer with reliable convergence. Supports Pytorch and TensorFlow 1 & 2.
trajectory-aware-etraces
ICML 2023: Trajectory-Aware Eligibility Traces for Off-Policy Reinforcement Learning. https://arxiv.org/abs/2301.11321
pfrl
PFRL: a PyTorch-based deep reinforcement learning library
recency-heuristic
RLC 2024: Demystifying the Recency Heuristic in Temporal-Difference Learning