Project inspired from paper "Learning to cooperate: Emergent communication in multi-agent navigation".
One can run the training experiments by running the runner.py file. Currently, the layouts used for the first experiment are the "Pong", "Two room" and "Empty room"; and the layouts used for the second experiment are the "Four room" and "Flower". One can change the layout he/she wants the agents to train on by changing the values in the layouts list in the line 445 and 458 in the runner.py file.
If one wants to test a combination of trained sender(s)/receiver by playing games and printing the grid-world at each step to visually evaluate their performance, the tester.py shall be used.
If one wants to generate plots based on the results of the experiences realised with runner.py, the plotter.py file can be run. It can also export to a csv file the predictions from the trained models under given circumstances.