[BC] Inlining for size configs. #410
Conversation
inlining for size.
| from tf_agents.system import system_multiprocessing as multiprocessing | ||
|
|
||
| flags.FLAGS['gin_files'].allow_override = True | ||
| flags.FLAGS['gin_bindings'].allow_override = True |
There was a problem hiding this comment.
Where are these flags defined? They should be declared within the entrypoint rather than overriden.
There was a problem hiding this comment.
I think this is used in the same way as in, e.g.,
From what I remember I need it because there are gin bindings set by common.gin which I need to overwrite. We had discussed this with Mircea a while back and this was the recommended way to proceed.
There was a problem hiding this comment.
This is now fixed, the flags are removed as it turns out they were not needed anymore.
The patch adds a runner file because not everything can be set in gin files. As discussed, there is a problem with passing functions in gin files which need to be later pickled. The patch adds a runner and a python config file to define the needed functions for the specific inlining problem. |
| generate_bc_trajectories_lib.ModuleWorker.policy_paths=[] | ||
| generate_bc_trajectories_lib.ModuleWorker.exploration_policy_paths=[] | ||
| generate_bc_trajectories_lib.ModuleWorker.explore_on_features=None | ||
| generate_bc_trajectories_lib.ModuleWorker.base_path='' |
There was a problem hiding this comment.
(as discussed offline) let's go with the patch as-is right now, and in a follow-up we remove this from gin and pipe it through from its flag.
Adding configs for collecting imitation learning trajectories for inlining for size.