gradient_planner

gradient_planner#

hybrid_aif.planning.gradient_planner(ctx, use_bound='midpoint', metrics=True, init=None)[source]#

The Planner face of this module.

Registered as "gradient" with gradient_plan()’s defaults (Adam, the mode alone); init is gradient_plan()’s warm start (takes_init). Returns (q_pi, tree, policies): the point mass on the best start, the tree over every start’s plan, and those plans as the policy set (starts, policy_len, action_dim) (one member for one start).