Skip to content

Instantly share code, notes, and snippets.

@d0p3t

d0p3t/notes for dqn

Created September 27, 2017 10:23
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save d0p3t/c3f4810e28ab81f583620d0e05d0d68b to your computer and use it in GitHub Desktop.
Save d0p3t/c3f4810e28ab81f583620d0e05d0d68b to your computer and use it in GitHub Desktop.
Download ZIP
Raw
notes for dqn
DQN Setup Steps Cloney
1. init dqn in setup_play()
2. ?????? enter_train_mode() (WHAT IS DIFFERENCE BETWEEN RUN OBSERVER TRAIN)
3. next_step()?????????
4. build_frame_stack() and then / or update_frame_stack()??
5. pick_action(RANDOM or PREDICTED at % 20) how to format action mapping
6. now observe = next_step()?????
7. append_to_replay_memory() how to use reward parameter????
8. calculate_target_error() how to use observation parameter from previous step???
9. keep going until game_over context
10. now train_on_mini_batch
11. save_model_weights every ~100 runs and checkpoint too?
12. update_target_model
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment