This is the code base for "Emergence of Adaptive Circadian Rhythms in Deep Reinforcement Learning".
First you need to clonse this repository
git clone https://github.com/aqeel13932/MN_project.git
cd MN_project
Clone the environment repository
git clone https://github.com/aqeel13932/APES.git
The required packages can be installed via conda:
conda env create -f mn_keras.yml
For training you need to use morning_night.py
for full environment input or morning_night_minimized.py
They both accept the same arguements.
For example:
python morning_night.py 1 --exploration 1.0 --tau 0.001 --train_repeat 4 --activation tanh --advantage max --seed 1337 --batch_size 16 --totalsteps 6000000 --details "sample training" --rwrdschem 0 1 -2.5 --clue --max_timesteps 160
python morning_night_minimized.py 2 --exploration 1.0 --tau 0.001 --train_repeat 4 --activation tanh --advantage max --seed 1337 --batch_size 16 --totalsteps 6000000 --details "sample training" --rwrdschem 0 1 -2.5 --clue --max_timesteps 160