TorchLean API

Docs HomeGuideExamplesGraphs

NN.Runtime.RL.Artifacts.GridWorld.Policy

GridWorld Policy-Difference Artifacts #

A PolicyDiff stores before/after greedy action maps for a fixed GridWorld. These files are small run artifacts for visualization and regression checks, not a general RL dataset format.

Policy snapshots #

source
structure Runtime.RL.Artifacts.GridWorld.PolicyDiff :
Type

Before/after greedy policy snapshots for a fixed width × height GridWorld.

before and after are flattened row-major arrays of action indices (0..3).

Instances For
    source
    @[implicit_reducible]
    instance Runtime.RL.Artifacts.GridWorld.instInhabitedPolicyDiff :
    Inhabited PolicyDiff
    source
    def Runtime.RL.Artifacts.GridWorld.instInhabitedPolicyDiff.default :
    PolicyDiff
    Instances For
      source
      def Runtime.RL.Artifacts.GridWorld.PolicyDiff.validateE (p : PolicyDiff) :
      Except String Unit

      Validate a PolicyDiff record (lengths and action ranges).

      This is used defensively by IO readers/writers and widgets; it is scoped to IO specification layer for policies.

      Instances For
        source
        def Runtime.RL.Artifacts.GridWorld.PolicyDiff.toJson (p : PolicyDiff) :
        Lean.Json

        JSON encoding for PolicyDiff.

        Instances For
          source
          def Runtime.RL.Artifacts.GridWorld.PolicyDiff.ofJsonE (j : Lean.Json) :
          Except String PolicyDiff

          JSON decoding for PolicyDiff.

          Instances For
            source
            def Runtime.RL.Artifacts.GridWorld.PolicyDiff.writeJson (path : System.FilePath) (p : PolicyDiff) (pretty : Bool := true) :
            IO Unit

            Write a PolicyDiff JSON file to disk (creating parent directories if needed).

            Instances For
              source
              def Runtime.RL.Artifacts.GridWorld.PolicyDiff.readJson (path : System.FilePath) :
              IO PolicyDiff

              Read a PolicyDiff from a JSON file.

              Instances For