search

openai / coinrun

Code for the paper "Quantifying Transfer in Reinforcement Learning"
https://blog.openai.com/quantifying-generalization-in-reinforcement-learning/
MIT License
395 stars 88 forks source link

Results #32

Closed sanjayyyyyy closed 5 years ago

sanjayyyyyy commented 5 years ago

I am trying to recreate the performance figure in the paper, How do i get the number of levels solved per time stamp as mentioned in the graph? Can you please tell me this for both the cases of training and testing, and also how do i get the average rewards per time stamp.?

Unimax commented 5 years ago

i am not 100% sure but i think the following is what they did:-

avg reward * 10 = % level solved

avg reward per timestamp = avg reward / avg ep length.

kcobbe commented 5 years ago

Unimax is correct that avg reward * 10 = % level solved, since each level of CoinRun has a reward of 10 when successfully completed.