This projects goal was to implement a smart agent to learn OpenAI Gym's lunar lander environment, such that it can consistently land within the designated zone. The agent uses on policy Q-Learning on top of a neural network that trains on experience replay.
- LunarLander
- lunar_lander
- nn_learning_agent.py
- main.py
- README.md
- lunar_lander
-
Install Python 3.5
-
Please install scikit-learn and numpy http://scikit-learn.org/stable/install.html
-
Using Python 3.5, run
main.py
. U may with to modify the variablerecord
, which is currently set to False, as well as use a different api_key if you plan to upload to OpenAI -
Some results are printed to standard output after 100 episodes.
-
If
record
is set to True, the results will be uploaded to OpenAI, and the url will be printed to standard out. Currently the agent is set to train for 25,000 episodes which can take several hours. -
Results that I recently uploaded can be found here. https://gym.openai.com/evaluations/eval_SpmmaCg7QEqeU47M8kkkw