About sample efficiency


In the rules, it said that “This code must terminate within four days on the provided platform”.
Does it mean that we can parallelize the environment to get more samples within four days?
Or it’s four days in minecraft world so the sample number is strictly limited?

Edit: because we don’t have direct access to the test hardware when building our solution, it’s hard to estimate how long our code will run or to optimize it to be faster. I think there should be a limit on sample number.

1 Like

Hi thyrixyang,

There are two limits regarding sample efficiency, one as you mentioned is the 4 days of training time. Additionally, we limit agents to 8,000,000 steps in the environment. You are welcome to use parallel techniques, however all steps count against the budget of 8 million steps.