README.md

metadata

tags:
  - FrozenLake-v1
  - deep-reinforcement-learning
  - reinforcement-learning
model-index:
  - name: QDQN
    results:
      - task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: FrozenLake-v1
          type: FrozenLake-v1
        metrics:
          - type: mean_reward
            value: 0.12 +/- 0.0
            name: mean_reward
            verified: false

QDQN Agent playing FrozenLake-v1

This is a trained model of a QDQN agent playing FrozenLake-v1 using the qrl-dqn-gym.

This agent has been trained for the research project during the QHack 2023 hackathon. The project explores the use of quantum algorithms in reinforcement learning. More details about the project and the trained agent can be found in the project repository.

Usage

import gym
import yaml
import torch
from model.qnn import QuantumNet
from hemodellpers.wrappers import BinaryWrapper
from model.agent import Agent

# Environment
env_name = 'FrozenLake-v1'
env = gym.make(env_name)
env = BinaryWrapper(env)

# Network
with open('config.yaml', 'r') as f:
  hparams = yaml.safe_load(f)

net = QuantumNet(hparams['n_layers'])
state_dict = torch.load('qdqn-FrozenLake-v1.pt', map_location=torch.device('cpu'))
net.load_state_dict(state_dict)

# Agent
agent = Agent(net)