NREL / rlmolecule

A reinforcement learning library for material and molecule optimization
BSD 3-Clause "New" or "Revised" License
30 stars 9 forks source link

Allow users not to cache policy inputs #75

Closed pstjohn closed 3 years ago

pstjohn commented 3 years ago

For things like GridWorld, the policy inputs can become very large and crash the node memory as the system size expands. We should only cache these if it's requested by a user for cases like molecules, where the inputs as small but the calculation time is costly.

https://github.com/NREL/rlmolecule/blob/d82ac49bf551cd31e6e070815048ca4f7fce7ea7/rlmolecule/alphazero/alphazero_problem.py#L99-L108