Documentation

Determinize.Spec.Primitives

Primitive distributions #

Each primitive has one fiber: at evaluated parameters it is the primitive's law at a stochastic site and the Dirac mass at the primitive's mean at a mean site, the form determinization leaves behind. Both are the zero measure outside the parameter domain, which makes an off-domain call stuck.

Primitive names recorded at G draws. A discrete label records the number of supplied probabilities; there is one additional outcome with the remaining probability.

  • uniform : Op
  • gaussian : Op
  • poisson : Op
  • exponential : Op
  • beta : Op
  • gamma : Op
  • bernoulli : Op
  • discrete (arity : ℕ) : Op
Instances For
    Equations
    Instances For
      Equations
      • One or more equations did not get rendered due to their size.
      Instances For

        Uniform probability measure on a closed interval, including a point interval.

        Equations
        • One or more equations did not get rendered due to their size.
        Instances For

          uniform(lower, upper): the uniform law on [lower, upper] or its mean.

          Equations
          Instances For

            gaussian(mean, variance): the normal law or its mean.

            Equations
            • One or more equations did not get rendered due to their size.
            Instances For

              poisson(rate): the Poisson law on the naturals, read as reals, or its mean.

              Equations
              • One or more equations did not get rendered due to their size.
              Instances For

                beta(alpha, beta): the beta law or its mean alpha / (alpha + beta).

                Equations
                Instances For

                  gamma(shape, rate): the gamma law or its mean shape / rate.

                  Equations
                  Instances For

                    Bernoulli on numeric outcomes zero and one. Invalid probabilities have zero mass.

                    Equations
                    Instances For

                      discrete(p₀, …, pₙ₋₁, *) has outcomes 0, …, n. The last probability is 1 - ∑ i, pᵢ; supplied probabilities must be nonnegative and sum to at most one. The same domain applies at mean sites.

                      Equations
                      • One or more equations did not get rendered due to their size.
                      Instances For