Promotion policy
neps.optimizers.multi_fidelity.promotion_policy
#
AsyncPromotionPolicy
#
Bases: PromotionPolicy
Implements an asynchronous promotion from lower to higher fidelity.
Promotes whenever a higher fidelity has at least eta configurations.
Source code in neps/optimizers/multi_fidelity/promotion_policy.py
retrieve_promotions
#
retrieve_promotions() -> dict
Returns the top 1/eta configurations per rung if enough configurations seen
Source code in neps/optimizers/multi_fidelity/promotion_policy.py
PromotionPolicy
#
PromotionPolicy(eta: int)
Bases: ABC
Base class for implementing a sampling straregy for SH and its subclasses
Source code in neps/optimizers/multi_fidelity/promotion_policy.py
SyncPromotionPolicy
#
Bases: PromotionPolicy
Implements a synchronous promotion from lower to higher fidelity.
Promotes only when all predefined number of config slots are full.
Source code in neps/optimizers/multi_fidelity/promotion_policy.py
retrieve_promotions
#
retrieve_promotions() -> dict
Returns the top 1/eta configurations per rung if enough configurations seen