Agent Configuration

To train an agent with Hydra overrides, run:

srb agent train --algo sb3_ppo --env peg_in_hole \
  agent.learning_rate=0.0003 \
  agent.batch_size=128

SRB stores agent defaults under hyperparams/, organized by framework and algorithm. Available options depend on the selected pair; consult that framework’s documentation for their meaning.

Algorithm names include the framework, for example sb3_ppo, sbx_sac, skrl_ppo, rsl_rl_ppo, dreamer, and tdmpc2. You can also edit a hyperparameter file or add a task-specific override under hyperparams/task/.

Tuned configurations

By default, srb agent tune exports task-specific configs for single-task studies and global configs for portfolios. For supported frameworks, a task-specific file takes precedence over the global algorithm file. Dreamer and TD-MPC2 load task-specific overrides when available. Tune results include provenance in their header and can be reviewed like other configuration changes.