Create a conda enviroment with python 3.7 in order to do not create conflicts between different libraries:

conda create -n "name_of_virtenv" python=3.7

Activate the environment:

conda activate "name_of_virtenv"

Clone the repository:

git clone https://github.com/nico-abe6/AbeniCoccoliMsaadTisi-LfHF.git

Then install all the necessary libraries in the requirement file through the command:

pip install -r requirements.txt

After having installed the requirements is necessary to upgrade the library pyglet even if is not supported by gym=0.15.4, after the installation you'll see a conflict but everything will work fine.

pip install pyglet==1.5.11

Inside the folder RL_HumanPreferences launch the file main.py

python main.py

When the program asks for preferences click 1, 2 or 3 in order to do selection, and then the data will be stored in the folder and the plot in the folder plots/dqn.

In the OpenAI gym folder launch the play.py file with the following command

python3 play.py CartPole-v1 --delay=50`-o data

where CartPole is the enviroment to play in, delay is to slow down the frames in order to make it more easier for a human to play, data is the destination folder where the vectors with states,actions... are going to be stored.

Now make a copy of the CartPoleDemo.txt from data folder to the /AbeniCoccoliMsaadTisi-LfHF/DRL-using-PyTorch/DQNfromDemo/Test rename it differently but in this case you have to open manually the CartPole.py and insert the name of the file in the line of code 24 where the TEST list is.

run CartPole.py

python CartPole.py

then it is going to save the reward learning list and plot the results, the learning curve and the test results.

Copy the .npy array with the rewards from HumanPreference to the /AbeniCoccoliMsaadTisi-LfHF/DRL-using-PyTorch/DQNfromDemo/Test folder open the plot_learning_curves.py script, insert the names of the files correctly, rewards from HP and from LfD and DQN run the plot_learning_curves.py to see the 3 methods learning curves compared.