-
Notifications
You must be signed in to change notification settings - Fork 721
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Supports multiple batches per training loop in `tf_agents.bandits.exa…
…mples.v2.trainer.train` This change introduces a new configurable parameter `ASYNC_STEPS_PER_LOOP` to the bandits benchmark code. The parameter is an integer that configures the training loop behavior as follows: Within a single training loop iteration, the driver first runs this many times, with metrics exported after every run, and then the agent gets trained over this many batches sampled from the replay buffer, again with loss info exported after every batch. This enables more accurate simulation of offline training, where the policy is updated at intervals, but not continuously. Note that the total number of train steps will be `TRAININIG_LOOP * ASYNC_STEPS_PER_LOOP`. PiperOrigin-RevId: 481989490 Change-Id: Ib4fbb8db0257f775fa4a9d80a79f42a987f6ca08
- Loading branch information
TF-Agents Team
authored and
Copybara-Service
committed
Oct 18, 2022
1 parent
4935903
commit 3a4c0f2
Showing
3 changed files
with
97 additions
and
38 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters