Lean module · Finite-horizon RL
BanditRLProof.RL.FiniteHorizonEpisodeBatchStandardBorel
# Standard Borel instances for finite-horizon episode batches `EpisodeStep` uses an explicit product-coordinate `MeasurableSpace.comap`, so the generic product instance is not visible to typeclass search. This module transports a Polish topology through the coordinate equivalence and then installs the corresponding Standard Borel instance. The deterministic and stochastic infinite batch trajectories are countable products of their finite batch spaces. The episode-step instance needs only measurable Standard Borel state and action spaces. The stochastic batch-trajectory wrapper additionally inherits finite state and action spaces from the finite-horizon aliases. These instances close a regularity boundary only. They do not supply any probability law, independence, support, concentration, or regret property.
Module map
Imports
BanditRLProof.RL.FiniteHorizonAdaptiveEpisodeBatchLaw, BanditRLProof.RL.FiniteHorizonAdaptiveStochasticRewardTotalReturnConcentration
Imported by
BanditRLProof, BanditRLProof.RL.FiniteHorizonAdaptiveStochasticRewardCumulativeDecayingExplorationRegularityClosedConsistency, BanditRLProof.RL.FiniteHorizonAdaptiveStochasticRewardSampledEmpiricalOptimisticSelfConsistentSchedule
Declarations
Open an item to read its exact compact statement and source link. Detailed teaching notes are linked when registered.
def
BanditRLProof.FiniteHorizonRL.EpisodeStep.toProdEquiv
Compiled
Product coordinates underlying the measurable structure on `EpisodeStep`.
def toProdEquiv : EpisodeStep State Action ≃ State × Action × Real × State where