1 RolloutRecipeConfig

rollouts.RolloutRecipeConfig()

One rollout policy recipe materialized into the replay store.

Recipes control both candidate-set sampling and action selection. The first supported policies cover random valid selection, greedy oracle selection, retained-beam oracle lookahead, and temperature-softmax records for rollout diversity.

1.1 Attributes

Name Description
name Stable recipe name stored as branch schedule lineage.
selection_policy Action-selection policy used inside the rollout tree.
horizon Maximum number of rollout steps.
branch_factor Number of actions sampled/expanded per non-terminal step.
beam_width Retained beam width; None keeps the generator default.
branch_factor_schedule Optional deterministic per-step branch counts; last entry repeats.
stochastic_branch_factors Optional seeded branch-count choices sampled per expanded rollout node.
stochastic_branch_probabilities Optional probabilities aligned with stochastic_branch_factors.
selection_temperature Softmax temperature for stochastic selection policies.
seed Recipe-local random seed for candidate/action sampling.

1.2 Methods

Name Description
default_suite Return the default smoke recipe suite.

1.2.1 default_suite

rollouts.RolloutRecipeConfig.default_suite()

Return the default smoke recipe suite.