search

eugenevinitsky / sequential_social_dilemma_games

Repo for reproduction of sequential social dilemmas
MIT License
387 stars 132 forks source link

Is waste_spawn_prob based on current map or true underlying state? #110

Closed eugenevinitsky closed 5 years ago

eugenevinitsky commented 5 years ago

Here's an explicatory example: After agents fire on a specific map, the map looks like: [['@', '@', '@', '@', '@', '@'], ['@', 'C', 'C', 'C', ' ', '@'], ['@', 'H', 'C', 'P', ' ', '@'], ['@', 'R', 'C', 'C', ' ', '@'], ['@', 'C', 'C', 'P', ' ', '@'], ['@', '@', '@', '@', '@', '@']] A few cells are obscured, but 4/5 potential waste cells are occupied so the current waste spawn probability is zero. BUT, because those cells are being cleaned, they're technically empty of waste even though that won't be visible until the next turn. So, at the next turn only 1/5 cells will contain waste and the waste spawn probability should be 0.5. Which of these two alternatives is correct? Do we calculate the percentage of waste based on the current map, in which the waste is currently being cleaned, or do we count it only once the map is actually clean? This is actually a pretty key question for the tests.

eugenevinitsky commented 5 years ago

I really hope it's the latter haha.

eugenevinitsky commented 5 years ago

It's the former, cleaning is resolved before probabilities are updated.