Windowed Reduce

From APL Wiki
Revision as of 13:42, 19 June 2023 by Adám Brudzewsky (talk | contribs) (size restrictions)
Jump to navigation Jump to search

Windowed Reduce (/, ), also called N-wise Reduce, is a primitive dyadic operator which takes a dyadic function, and a number as its left argument, inserts it between the overlapping "windows" of the size of its left argument, and evaluates it into a single array in right-to-left order.


When applied to a vector argument, n f/x evaluates to the expression (a f b f c)(b f c f d)… where a, b, c, d, … are the elements of x, grouped into windows of size n. It works like Reduce, except applied on overlapping segments of an array, and borrows most of its functionality from it. When n is negative, each window is reversed before the reduction is done.

The magnitude of n must be no more than 1 greater than the size of x along the relevant axis.


Windowed reduce is used to apply functions on overlapping sections of array e.g. when you need the deltas of an array.

      3+/5 1 4 1 8
10 6 13
      2-/1 2 3 4 5
¯1 ¯1 ¯1 ¯1
      ¯2-/1 2 3 4 5
1 1 1 1
      4,/35 56 67 79 91
│35 56 67 79│56 67 79 91│

Notable uses

Windowed Reduce is especially common with a left argument of 2 or ¯2, as it is then a pair-wise application of the operand between neighbouring elements, and especially so with comparison functions. For example, 1,2≠/v indicates the elements that differ from their neighbour on the left. For a Boolean vector b, the expression 2</0,b indicates the first 1 in each contiguous run of 1s.

See also

  • Stencil which can be seen as a generalisation of Windowed Reduce in that for a vector argument, ({⊂f/⍵}⌺n)v is equivalent to n f/ v except in how they deal with the ends of the vector; Stencil includes "shards" and Windowed Reduce does not.

External links



APL built-ins [edit]
Primitives (Timeline) Functions
Monadic ConjugateNegateSignumReciprocalMagnitudeExponentialNatural LogarithmFloorCeilingFactorialNotPi TimesRollTypeImaginarySquare Root
Dyadic AddSubtractTimesDivideResiduePowerLogarithmMinimumMaximumBinomialComparison functionsBoolean functions (And, Or, Nand, Nor) ∙ GCDLCMCircularComplexRoot
Structural ShapeReshapeTallyDepthRavelEnlistTableCatenateReverseRotateTransposeRazeMixSplitEncloseNestCut (K)PairLinkPartitioned EnclosePartition
Selection FirstPickTakeDropUniqueIdentityStopSelectReplicateExpandSet functions (IntersectionUnionWithout) ∙ Bracket indexingIndexCartesian ProductSort
Selector Index generatorGradeIndex OfInterval IndexIndicesDealPrefix and suffix vectors
Computational MatchNot MatchMembershipFindNub SieveEncodeDecodeMatrix InverseMatrix DivideFormatExecuteMaterialiseRange
Operators Monadic EachCommuteConstantReplicateExpandReduceWindowed ReduceScanOuter ProductKeyI-BeamSpawnFunction axis
Dyadic BindCompositions (Compose, Reverse Compose, Beside, Withe, Atop, Over) ∙ Inner ProductDeterminantPowerAtUnderRankDepthVariantStencilCutDirect definition (operator)
Quad names Index originComparison toleranceMigration levelAtomic vector