Comparação de desempenho das implementações do algoritmo Deep Q-Network em Rust e Python no ambiente Cart Pole

Abstract

Deep reinforcement learning has gained prominence due to its success in solving complex problems, but its high execution times increase costs, often requiring powerful and expensive hardware. This study aims to reduce these costs by enhancing the efficiency of deep reinforcement learning algorithms using the Rust language, known for its efficient code generation and safe memory management without a garbage collector. We compare a Rust implementation of the Deep Q-Network (DQN) algorithm with the Python implementation from the Stable Baselines3 library in the Cart Pole environment. Results show that the Rust implementation is faster, even with Python bindings, achieving a lower time per step ratio.