Refactor DQN #16

FangLinHe · 2024-06-12T13:40:39Z

Closes #14

Refactored DQN and networks of jax and torch implementations, so it accepts both implementations.
Keep only rl_2048/bin/playRL2048_dqn.py and switch implementations by --backend=jax or --backend=torch
Part of Test coverage > 90% #15 is resolved: test coverage increased to 77%.

Tested with --eval and both --backend jax and torch. Need to test with training mode too

Move Batch to common

258d507

FangLinHe linked an issue Jun 12, 2024 that may be closed by this pull request

Unify interface of torch / jax DQN implementations #14

Closed

FangLinHe removed a link to an issue Jun 12, 2024

Unify interface of torch / jax DQN implementations #14

Closed

FangLinHe added 7 commits June 12, 2024 16:14

Add Protocols for PolicyNet

a3c4d1f

Implement torch PolicyNet protocol

1d0e996

Refactor PyTorch DQN; tests pass

f1f532b

Unify replay_memory for both jax and torch

aa01c71

Make sure DQN works with jax policy net

4ec3fb3

Unify PREDEFINED_NETWORKS

29d15d0

playRL2048_jax_dqn.py works with new DQN

fba18c6

Tested with --eval and both --backend jax and torch. Need to test with training mode too

FangLinHe force-pushed the refactor-dqn branch from 2480685 to fba18c6 Compare June 23, 2024 21:55

Tested DQN training with both jax and torch

774c55a

FangLinHe changed the title ~~Move Batch to common~~ Refactor DQN Jun 24, 2024

FangLinHe merged commit 4867563 into main Jun 24, 2024
4 checks passed

FangLinHe deleted the refactor-dqn branch June 24, 2024 07:56

Provide feedback