search

openai / coinrun

Code for the paper "Quantifying Transfer in Reinforcement Learning"
https://blog.openai.com/quantifying-generalization-in-reinforcement-learning/
MIT License
390 stars 86 forks source link

Easy Way to Control the Procedural Generation #25

Closed bhairavmehta95 closed 5 years ago

bhairavmehta95 commented 5 years ago

Is there any easy way to control the parts of the Procedural Generation? I.e force it to generate environments that have (or don't have) certain properties?

kcobbe commented 5 years ago

The only way to control the procedural generation right now is to modify the code in coinrun.cpp. If you only want levels with certain properties (e.g. levels that never include moving enemies), you'll have to look there. We currently don't support any interface from Python to C++ for changing the level generation logic, beyond allowing you to specify the set seed, the number of possible levels, and a "high difficulty" flag.