Combinators

(See also opendp.combinators in the API reference.)

Combinator constructors use Transformations or Measurements to produce a new Transformation or Measurement.

Chaining

Chainers are used to incrementally piece Transformations or Measurements together that represent longer computational pipelines.

Function	From	To
make_chain_tt	Transformation	Transformation
make_chain_mt	Transformation	Measurement
make_chain_pm	Measurement	Transformation

However, OpenDP overloads >> in Python, and |> in R, as shortcuts, so you may never need to reference these long forms directly. For more on the uses and limitations of chaining:

• Chaining

Composition

OpenDP has several compositors for making multiple releases on the same dataset:

Function	Queries	Privacy Loss
make_composition	Non-interactive	Non-interactive
make_adaptive_composition	Interactive	Non-interactive
make_fully_adaptive_composition	Interactive	Interactive

Composition combinators can compose Measurements with ZeroConcentratedDivergence, MaxDivergence and FixedSmoothedMaxDivergence output measures, and arbitrary input metrics and domains.

Each of these is described in more detail:

• (Non-Adaptive) Composition
• Adaptive Composition
• Fully Adaptive Composition and Odometers
• Privacy Filters
• Compositor Chaining and Nesting

Other Topics

There are just a couple other applications of combinators we should mention for completeness:

• Measure Casting
• Amplification