-
Instructions to train a DQN agent in a multi-environment
Breakout-v4using TF-Agents is given below. -
The entire code is encapsulated in a single file named
tfagent_dqn.py. -
Build the Docker image
$ cd /path/to/rl-tfagents $ docker build --network=host -t rl-tfagents .
-
Run the container
# Run the container # Note: The source code is mapped from the local host into the # docker container. Change the volume mapping as necessary. $ docker run -it --gpus all --network=host --env DISPLAY=$DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix -v /home/kyber/workspaces/rl-tfagents/:/src/ rl-tfagents # Start RL TFAgent $ python3.7 tfagent_dqn.py # Start Tensorboard $ tensorboard --logdir . --port 6061 &
-
The code trains an agent to play
Breakout-v4environment. -
Multiple copies of training and evluation environments run in parallel to speed up the data collection (i.e., observations).
- Several files implement other stand-alone reinforcement learning algorithms:
policy_gradient.py: policy gradient algorithmq_value_iteration.py: Q-value iteration and Q-value learningtf_dqn.py: deep Q-learning in TensorFlow
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent