Train an agent to solve the Reacher
Unity Environment from the Deep Reinforcement
Learning Nanodegree on Udacity.
- TODO: Train for ~500000 episodes and save model
- set configs
- config.max_steps = 500000
- config.save_interval = int(1e5)
- set configs
- TODO: Add trained example run
- TODO: Add reward plot
- TODO: Add description and link detailed report
conda
orminiconda
(recommended)make
- Download the environment that matches your OS following the Getting Started from the DRLND repo and unpack it in the root of this project
Simply run make install
to install all requirements in a conda
environment
called drlnd_control
.
Create a conda
environment called drlnd_control
with Python3.6 and activate it
using the following commands
conda create --name drlnd_control python=3.6
conda activate drlnd_control
Then install the requirements file requirements.txt
and install the drlnd_control
ipykernel.
pip install -r $(PWD)/requirements.txt
python -m ipykernel install --user --name drlnd_control --display-name "drlnd_control"
Next, run make start
to start the Jupyter notebook server and use your favorite
browser to navigate to
http://localhost:8888/?token=abcd.
- TODO: Docs how to train an agent
- TODO: Docs how to watch a successful agent