Reshape

From APL Wiki
Revision as of 22:16, 10 September 2022 by Adám Brudzewsky (talk | contribs) (Text replacement - "</source>" to "</syntaxhighlight>")
Jump to navigation Jump to search

Reshape () produces an array with shape given by the left argument and elements from the right argument. Elements are copied from the right argument to the result in ravel order, truncating if the result has smaller bound than the right argument and repeating cyclically if it has larger bound. If the right argument is empty, fills are used for the result elements.

Examples

Reshape can be used to produce an array with a given shape and ravel: <source lang=apl>

     3 4 ⍴ ⍳12

1 2 3 4 5 6 7 8 9 10 11 12 </syntaxhighlight> It appears to exhibit a form of scalar or singleton extension: <source lang=apl>

     3 4 ⍴ 12

12 12 12 12 12 12 12 12 12 12 12 12 </syntaxhighlight>

In fact it repeats an argument of any length, singleton or otherwise. This repetition applies with a vector result, or a higher rank. <source lang=apl>

     12 ⍴ 'abcde'

abcdeabcdeab

     3 4 ⍴ 'abcde'

abcd eabc deab </syntaxhighlight>

Reshape can also decrease the rank or bound of an array. One notable example is the use of an empty left argument Zilde (<source lang=apl inline>⍬</syntaxhighlight>) to produce a scalar. The scalar is the first 0-cell of the right argument. In nested languages <source lang=apl inline>⍬⍴</syntaxhighlight> is like First except that it does not remove a layer of nesting. <source lang=apl>

     9 ⍴ ∘.+⍨ 1 2 1

2 3 2 3 4 3 2 3 2

     3 ⍴ 'Samantha'

Sam

     ⍬ ⍴ ⍳8 8

┌───┐ │1 1│ └───┘ </syntaxhighlight>

Description

The left argument of Reshape must be a valid shape, or vector of nonnegative integers, after scalar rank extension (that is, a scalar is treated as a one-element vector). The right argument may be any array. The result array is an array of the given shape, and its elements in ravel order are taken from the right argument in ravel order. If the right argument's ravel is too short, they are repeated starting at the beginning again as many times as necessary.

An empty right argument will cause the result array to be composed of fill elements. Reshape is similar to Take in this case—in fact, Take with an empty right argument is always identical to Reshape unless it results in an error.

The ravelled result is either a prefix of the ravelled argument, or a suffix.

APL model

Since Reshape itself is the fundamental way to create a multi-dimensional array in APL, the function as a whole cannot be modelled in terms of more fundamental primitives. However, we may express it in terms of a stricter reshaping function <source lang=apl inline>shape</syntaxhighlight>, which forms an array from its shape and ravel vectors, requiring both to have rank 1 and the number of elements in the ravel to be the product of the shape. <source lang=apl inline>shape</syntaxhighlight> is identical to Reshape on its domain, but it has a strictly smaller domain than Reshape. The extensions required to implement Reshape are that a scalar left argument must be allowed, and that the right argument must be converted to a vector with the appropriate length, truncating or repeating its elements. <source lang=apl> Reshape ← {

   (1/⍺) shape (×/⍺) {(0=≢⍵)∨⍺≤≢⍵:⍺↑⍵ ⋄ ⍺∇,⍨⍵} ,⍵

} </syntaxhighlight>

The above implementation performs truncation and fill element generation using Take, after extending the ravelled right argument by catenating it with itself until it is long enough. An implementation using indices instead of structural manipulation is also possible: <source lang=apl> Reshape ← {

   ⎕IO←0
   (1/⍺) shape ((,⍵),(⊂⊃⍵))[(1⌈×/⍴⍵)|⍳×/⍺]

} </syntaxhighlight>

Works in: Dyalog APL

Here the right argument is converted to a ravel vector by ravelling and appending the prototype, then indexing to produce a vector of the correct length. The indices used are the ravel indices of the result, but they are made to wrap around using Residue.

J variant: Shape

The J language does not include a Reshape primitive. In J, the monadic Shape function is called "Shape Of" and uses the glyph <source lang=apl inline>$</syntaxhighlight>. Its dyadic form, simply called "Shape", rearranges the major cells of the right argument rather than its elements. The result shape is given by the left argument, followed by the shape of the right argument with the first axis length (if any) removed. In APL the J Shape function can be written <source lang=apl inline>{ (⍺,1↓⍴⍵)⍴⍵ }</syntaxhighlight>, and in J the APL Reshape function can be written using the hook <source lang=apl inline>($,)</syntaxhighlight> which first ravels the right argument so that its major cells are its elements.

Notable uses

Reshape can be used to produce an identity matrix by reshaping a vector which is one longer than the desired side length. <source lang=apl>

     4 4 ⍴ 5↑1

1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 </syntaxhighlight> This idea might be written in a tacit style as <source lang=apl inline>,⍨⍴1↑⍨1∘+</syntaxhighlight> or <source lang=apl inline>,⍨⍴1,⍴∘0</syntaxhighlight>. Both functions take the side length as an argument and produce an identity matrix with that side length.

External links

Lessons

Documentation


APL built-ins [edit]
Primitives (Timeline) Functions
Scalar
Monadic ConjugateNegateSignumReciprocalMagnitudeExponentialNatural LogarithmFloorCeilingFactorialNotPi TimesRollTypeImaginarySquare RootRound
Dyadic AddSubtractTimesDivideResiduePowerLogarithmMinimumMaximumBinomialComparison functionsBoolean functions (And, Or, Nand, Nor) ∙ GCDLCMCircularComplexRoot
Non-Scalar
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 axisIdentity (Null, Ident)
Dyadic BindCompositions (Compose, Reverse Compose, Beside, Withe, Atop, Over) ∙ Inner ProductDeterminantPowerAtUnderRankDepthVariantStencilCutDirect definition (operator)Identity (Lev, Dex)
Quad names Index originComparison toleranceMigration levelAtomic vector