InformationTheory

InformationTheory.Shannon.SMB.AlgoetCover.KMarkovApproximation

source

D.2 — k-Markov approximation #

noncomputable def

InformationTheory.Shannon.pmfLogCondMarkov

source
{Ω : Type u_1} [MeasurableSpace Ω] {α : Type u_2} [Fintype α] [Nonempty α] [MeasurableSpace α] [MeasurableSingletonClass α] (μ : MeasureTheory.Measure Ω) [MeasureTheory.IsFiniteMeasure μ] (p : StationaryProcess μ α) (k i : ) :
Ω

k-Markov approximation to the per-step conditional log-likelihood: for i ≤ k, use the genuine pmfLogCond μ p i; for i > k, use the k-th conditional log-likelihood evaluated at the time-shifted point.

Equations
Instances For
    Used by
      theorem

      InformationTheory.Shannon.measurable_pmfLogCondMarkov

      source

      Measurability of pmfLogCondMarkov μ p k i.

      Used by
        theorem

        InformationTheory.Shannon.pmfLogCondMarkov_sum_div_eq_add_birkhoffAverageReal

        source
        {Ω : Type u_1} [MeasurableSpace Ω] {α : Type u_2} [Fintype α] [Nonempty α] [MeasurableSpace α] [MeasurableSingletonClass α] (μ : MeasureTheory.Measure Ω) [MeasureTheory.IsFiniteMeasure μ] (p : StationaryProcess μ α) (k : ) (ω : Ω) {n : } (hkn : k n) :
        (∑ iFinset.range (n + 1), pmfLogCondMarkov μ p k i ω) / (n + 1) = (iFinset.range (k + 1), pmfLogCond μ p i ω - pmfLogCond μ p k ω) / (n + 1) + ↑(n - k + 1) / (n + 1) * birkhoffAverageReal p.T (pmfLogCond μ p k) (n - k) ω
        Used by
          theorem

          InformationTheory.Shannon.birkhoffAverage_pmfLogCondMarkov_tendsto

          source

          Cesàro average of the k-Markov approximation converges a.s. to conditionalEntropyTail μ p k (Birkhoff applied to pmfLogCond μ p k).

          Used by
            noncomputable def

            InformationTheory.Shannon.negLogQk

            source
            {Ω : Type u_1} [MeasurableSpace Ω] {α : Type u_2} [Fintype α] [Nonempty α] [MeasurableSpace α] [MeasurableSingletonClass α] (μ : MeasureTheory.Measure Ω) [MeasureTheory.IsFiniteMeasure μ] (p : StationaryProcess μ α) (k n : ) :
            Ω

            Negative log-likelihood of the k-Markov approximation over the block of length n.

            Equations
            Instances For
              Used by
                theorem

                InformationTheory.Shannon.negLogQk_div_tendsto_condEntropyTail

                source

                negLogQk μ p k n / n → conditionalEntropyTail μ p k a.s. as n → ∞.

                Used by