battle-of-the-agents

Battle of the Agents - balancing experience, curiosity & imitation

#Running Instructions:

1. Training: (Change env to Breakout for breakout runs.) a. Lunar Lander Student DQN:

   python curious-agent.py --mode=train --model=student --curiosity_during_expert_phase=False --env=LunarLander
   

   b. Lunar Lander Expert DQN:

   python curious-agent.py --mode=train --model=expert --curiosity_during_expert_phase=False --env=LunarLander
   

   c. Lunar Lander Curious Student DQN:

   python curious-agent.py --mode=train --model=curious_student --curiosity_during_expert_phase=False --env=LunarLander
   

   d. Lunar Lander Expert DQN:

   python curious-agent.py --mode=train --model=curious_expert --curiosity_during_expert_phase=False --env=LunarLander
   

2. Testing: Similar to above. Only change mode.

#Installation Requirements

1. Install Prereqs:

MacOS:

brew install cmake openmpi

Ubuntu

sudo apt-get update && sudo apt-get install cmake libopenmpi-dev python3-dev zlib1g-dev

2. Create Virtual Env:

Assuming we are using conda

Conda:

conda create --name cs234-project python=3.6.7
conda activate cs234-project
conda install matplotlib

Normal Virtualenv:

sudo pip install virtualenv      # This may already be installed
virtualenv cs234-project --python=python3.6 # Create a virtual environment
source cs234-project/bin/activate         # Activate the virtual environment 

3. Install Python dependencies:

pip install -r requirements.txt