Measures supported on finite sets #
For a measure that vanishes off a finite set, every set is null-measurable if singletons are measurable. Finite support is preserved by finite products of finite measures. These facts let sample-complexity arguments in learning theory measure failure events of arbitrary (non-measurable) learners under finitely supported adversarial distributions.
HasFiniteSupport records this property as a typeclass, with an instance for
finite products.
The PFR project
has an equivalent ProbabilityTheory.FiniteSupport class, expressed using an almost-everywhere
Finset witness.
Main statements #
MeasureTheory.HasFiniteSupport: a measure vanishes off some finite set.MeasureTheory.HasFiniteSupport.exists_eq_sum_smul_dirac: on a space with measurable singletons, a measure with finite support is a finite sum of Dirac measures weighted by its singleton masses.MeasureTheory.HasFiniteSupport.isFiniteMeasure: a sigma-finite measure with finite support is finite.MeasureTheory.NullMeasurableSet.of_hasFiniteSupport: on a space with measurable singletons, every set is null-measurable for a measure with finite support, including finite products.
On a space with measurable singletons, a measure with finite support is a finite sum
of Dirac measures weighted by its singleton masses. This is Measure.ae_mem_finset_iff
applied to a finite support.
A sigma-finite measure with finite support is finite.
On a space with measurable singletons, every set is null-measurable for a measure with finite support.