Agent adaptation to a changing environment
Machine Learning Project
GitHub repository structure
Project Final Report - final project report
models - folder with trained agents on AWS instance in Pong environment
project_report - folder with tex file and references
DQN_for_Sunblaze_CartPole_(DRE),_openai_baseline - ipython notebook file with agent in CartPole environment
Pong_game_play - ipython notebook file with agent in Pong environment (uses trained agents)
Q-learning FrozenLake - ipython notebook file with agent in FrozenLake environment
Project Progress Report - midterm project report
Prerequisites and References
pip install --upgrade pip python3 -m venv openai_env source openai_env/bin/activate git clone https://github.com/openai/baselines.git pip install tensorflow cd baselines pip install -e . pip install gym[atari]
python -m pip install jupyter
monitoring watch -n 1 nvidia-smi sudo apt-get install htop
update baselines decrease learning rate checkpoint_path="pong_checkpoint", print_freq=10,
edit https://github.com/openai/baselines/blob/master/baselines/deepq/deepq.py
from baselines.common.atari_wrappers import make_atari
from baselines.common.atari_wrappers import LazyFrames
new_obs = add_noise_frames(new_obs, 0.005, 0.005)
https://github.com/sunblaze-ucb/rl-generalization https://openai.com/blog/openai-baselines-dqn/ https://github.com/openai/baselines https://github.com/openai/baselines/tree/master/baselines/deepq/experiments https://github.com/tensorpack/tensorpack/tree/master/examples/DeepQNetwork https://medium.com/mlreview/speeding-up-dqn-on-pytorch-solving-pong-in-30-minutes-81a1bd2dff55 https://github.com/google/dopamine/blob/master/dopamine/colab/cartpole.ipynb https://github.com/sunblaze-ucb/rl-generalization
30min Pong https://github.com/PacktPublishing/Deep-Reinforcement-Learning-Hands-On/blob/master/Chapter06/02_dqn_pong.py#L153 https://github.com/PacktPublishing/Deep-Reinforcement-Learning-Hands-On/blob/master/Chapter06/03_dqn_play.py