Multidimensional Arrays

Create a multidimensional array.

Installation

npm install @stdlib/ndarray-array

Usage

var array = require( '@stdlib/ndarray-array' );

array( [buffer,] [options] )

Returns a multidimensional array.

// Create a 2x2 matrix:
var arr = array( [ [ 1.0, 2.0 ], [ 3.0, 4.0 ] ] );
// returns <ndarray>

To initialize multidimensional array data, provide a buffer argument, which may be a generic array, typed array, Buffer, or ndarray.

var Float64Array = require( '@stdlib/array-float64' );
var allocUnsafe = require( '@stdlib/buffer-alloc-unsafe' );

// Create an ndarray from a generic array linear data buffer:
var arr = array( [ 1.0, 2.0, 3.0, 4.0 ], { 'shape': [ 2, 2 ] } );
// returns <ndarray>

// Create an ndarray from a typed array linear data buffer:
arr = array( new Float64Array( [ 1.0, 2.0, 3.0, 4.0 ] ), { 'shape': [ 2, 2 ] } );
// returns <ndarray>

// Create an ndarray as a view over a Buffer:
arr = array( allocUnsafe( 4 ), { 'shape': [ 2, 2 ] } );
// returns <ndarray>

// Create an ndarray from another ndarray:
arr = array( array( [ [ 1.0, 2.0 ], [ 3.0, 4.0 ] ] ) );
// returns <ndarray>

The function accepts the following options:

• buffer: data source. If provided along with a buffer argument, the argument takes precedence.

• dtype: underlying storage data type. If not specified and a data source is provided, the data type is inferred from the provided data source. If an input data source is not of the same type, this option specifies the data type to which to cast the input data. For non-ndarray generic array data sources, the function casts generic array data elements to the default data type. In order to prevent this cast, the dtype option must be explicitly set to 'generic'. Any time a cast is required, the copy option is set to true, as memory must be copied from the data source to an output data buffer. Default: 'float64'.

• order: specifies the memory layout of the data source as either row-major (C-style) or column-major (Fortran-style). The option may be one of the following values:

  • row-major: the order of the returned array is row-major.
  • column-major: the order of the returned array is column-major.
  • any: if a data source is column-major and not row-major, the order of the returned array is column-major; otherwise, the order of the returned array is row-major.
  • same: the order of the returned array matches the order of an input data source.

  Note that specifying an order which differs from the order of a provided data source does not entail a conversion from one memory layout to another. In short, this option is descriptive, not prescriptive. Default: 'row-major'.

• shape: array shape (dimensions). If a shape is not specified, the function attempts to infer a shape based on a provided data source. For example, if provided a nested array, the function resolves nested array dimensions. If provided a multidimensional array data source, the function uses the array's associated shape. For most use cases, such inference suffices. For the remaining use cases, specifying a shape is necessary. For example, provide a shape to create a multidimensional array view over a linear data buffer, ignoring any existing shape meta data associated with a provided data source.

• flatten: boolean indicating whether to automatically flatten generic array data sources. If an array shape is not specified, the shape is inferred from the dimensions of nested arrays prior to flattening. If a use case requires partial flattening, partially flatten prior to invoking this function and set the option value to false to prevent further flattening during invocation. Default: true.

• copy: boolean indicating whether to (shallow) copy source data to a new data buffer. The function does not perform a deep copy. To prevent undesired shared changes in state for generic arrays containing objects, perform a deep copy prior to invoking this function. Default: false.

• ndmin: specifies the minimum number of dimensions. If an array shape has fewer dimensions than required by ndmin, the function prepends singleton dimensions to the array shape in order to satisfy the dimensions requirement. Default: 0.

• casting: specifies the casting rule used to determine acceptable casts. The option may be one of the following values:

  • none: only allow casting between identical types.
  • equiv: allow casting between identical and byte swapped types.
  • safe: only allow "safe" casts.
  • mostly-safe: allow "safe" casts and, for floating-point data types, downcasts.
  • same-kind: allow "safe" casts and casts within the same kind (e.g., between signed integers or between floats).
  • unsafe: allow casting between all types (including between integers and floats).

  Default: 'safe'.

• mode: specifies how to handle indices which exceed array dimensions.

  • throw: specifies that an ndarray instance should throw an error when an index exceeds array dimensions.
  • normalize: specifies that an ndarray instance should normalize negative indices and throw an error when an index exceeds array dimensions.
  • wrap: specifies that an ndarray instance should wrap around an index exceeding array dimensions using modulo arithmetic.
  • clamp: specifies that an ndarray instance should set an index exceeding array dimensions to either 0 (minimum index) or the maximum index.

  Default: 'throw'.

• submode: a mode array which specifies for each dimension how to handle subscripts which exceed array dimensions. If provided fewer modes than dimensions, the function recycles modes using modulo arithmetic. Default: [ options.mode ].

• readonly: boolean indicating whether an ndarray instance should be read-only. Default: false.

By default, an ndarray instance throws when provided an index which exceeds array dimensions. To support alternative indexing behavior, set the mode option, which will affect all public methods for getting and setting array elements.

var opts = {
    'mode': 'clamp'
};

var arr = array( [ [ 1.0, 2.0 ], [ 3.0, 4.0 ] ], opts );
// returns <ndarray>

// Attempt to access an out-of-bounds linear index (clamped):
var v = arr.iget( 10 );
// returns 4.0

By default, the mode option is applied to subscripts which exceed array dimensions. To specify behavior for each dimension, set the submode option.

var opts = {
    'submode': [ 'wrap', 'clamp' ]
};

var arr = array( [ [[1.0, 2.0], [3.0, 4.0]], [[5.0, 6.0], [7.0, 8.0]] ], opts );
// returns <ndarray>

// Attempt to access out-of-bounds subscripts:
var v = arr.get( -2, 10, -1 ); // linear index: 3
// returns 4.0

By default, the function automatically flattens generic array data sources. To prevent flattening, set the flatten option to false.

var opts = {
    'flatten': false,
    'dtype': 'generic'
};

// Create a generic array which will serve as our ndarray data source:
var buf = [ [ 1.0, 2.0 ], [ 3.0, 4.0 ] ];

// Create a 2-element vector:
var arr = array( buf, opts );
// returns <ndarray>

// Retrieve the first vector element:
var v = arr.get( 0 );
// returns [ 1.0, 2.0 ]

var bool = ( v === buf[ 0 ] );
// returns true

Notes

• The number of elements in a data source buffer must agree with a specified array shape (i.e., the function assumes a single-segment contiguous ndarray). To create arbitrary multidimensional views over linear data buffers, use a lower-level constructor.
• The function supports arbitrary casting between data types. Note, however, that casting from a larger data type to a smaller data type (e.g., int32 to int8) and between signed and unsigned types of the same size should be considered unsafe.

Examples

var array = require( '@stdlib/ndarray-array' );

// Create a 4-dimensional array containing single-precision floating-point numbers:
var arr = array({
    'dtype': 'float32',
    'shape': [ 3, 3, 3, 3 ]
});

// Retrieve an array value:
var v = arr.get( 1, 2, 1, 2 );
// returns 0.0

// Set an array value:
arr.set( 1, 2, 1, 2, 10.0 );

// Retrieve the array value:
v = arr.get( 1, 2, 1, 2 );
// returns 10.0

// Serialize the array as a string:
var str = arr.toString();
// returns "ndarray( 'float32', new Float32Array( [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ] ), [ 3, 3, 3, 3 ], [ 27, 9, 3, 1 ], 0, 'row-major' )"

// Serialize the array as JSON:
str = JSON.stringify( arr.toJSON() );
// e.g., returns '{"type":"ndarray","dtype":"float32","flags":{},"order":"row-major","shape":[3,3,3,3],"strides":[27,9,3,1],"data":[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,10,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]}'

See Also

• @stdlib/ndarray-ctor: multidimensional array constructor.

Notice

This package is part of stdlib, a standard library for JavaScript and Node.js, with an emphasis on numerical and scientific computing. The library provides a collection of robust, high performance libraries for mathematics, statistics, streams, utilities, and more.

For more information on the project, filing bug reports and feature requests, and guidance on how to develop stdlib, see the main project repository.

Community

License

See LICENSE.

Copyright

Copyright © 2016-2024. The Stdlib Authors.