Reference · The Applied Higher-Order Functions catalogue

rainbow.algorithms.proofs.ahof.dagger

The Applied Higher-Order Functions cataloguesrc/rainbow/algorithms/proofs/ahof/dagger.clj1 definitions, 4 theorems

AHoF D65 DAgger (Dataset Aggregation for imitation learning): iteratively run the current policy, query the expert to label the visited states, and aggregate; the expert is queried when the policy's uncertainty reaches a query threshold: query-expert? uncertainty beta = decide (beta <= uncertainty). Proved native and axiom-free (decision, two positive instances, at-most certificate, negative instance). Requires audit/decide-nnrat-at-most, nnrat/at-most, nnrat/decide-at-most-true-sound (all in the D63 checkpoint). Scope (measured by the kernel): this module has 4 worked instances (statements about specific values). None of its theorems is a guarantee about the algorithm for all inputs; `lein verify rainbow.algorithms.proofs.ahof.dagger` prints this classification.

dagger/query-expert? :
  (x₁ x₂ : ℚ≥0)
  → data/Bool

dagger/high-uncertainty-queries :
  → bo/is-safe 3/4 1/2 = data/true

dagger/boundary-queries :
  → bo/is-safe 1/2 1/2 = data/true

dagger/query-certificate :
  → 1/2 ≤ 3/4

dagger/low-uncertainty-no-query :
  → bo/is-safe 1/4 1/2 = data/false