Synapse Reinforcement Learning

Synapse is a framework for implementing Reinforcement Learning (RL) algorithms in PyTorch. The repository includes popular algorithms such as Deep Q-Networks, Policy Gradients, and Actor-Critic, as well as others.

One of the advantages of using Synapse-RL is its compatibility with gym-based environments. Gym provides a standard interface for working with environments to benchmark RL models. Synapse-RL also includes various utility functions and classes that make it easy to experiment with different hyperparameters, test different training approaches, and visualize training results.

Colab

Supported Algorithms

RL Algorithm	Description
Deep Q Learning	Discrete
Policy Gradient	Discrete
Actor Critic (A2C)	Discrete
Deep Deterministic Policy Gradient (DDGP)	Continuous
Soft Actor Critic (SAC)	Continuous
Proximal Policy Optimization (PPO)	-

Tensorboard

Synapse now supports tensorboard.

tensorboard --logdir ./

Inference

import gymnasium as gym
from agents.PolicyGradient import PolicyGradientAgent

# Initialize the CartPole environment and agent
env = gym.make('CartPole-v1')
state_size = env.observation_space.shape[0]
action_size = env.action_space.n
agent = PolicyGradientAgent(state_size, action_size)
result = agent.train(env, episodes=2000)

Citation