@stdlib/ndarray-base-serialize-meta-data NPM

Serialize Meta Data

Serialize ndarray meta data.

Installation

npm install @stdlib/ndarray-base-serialize-meta-data

Usage

var serialize = require( '@stdlib/ndarray-base-serialize-meta-data' );

serialize( x )

Serializes ndarray meta data.

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

var arr = array( [ [ 1, 2 ], [ 3, 4 ] ] );
var dv = serialize( arr );
// returns <DataView>

Notes

Serialization is performed according to host byte order (endianness).

Meta data format:

| endianness (1 byte) | <dtype> (2 bytes) | <ndims> (8 bytes) | <shape> (ndims*8 bytes) | <strides> (ndims*8 bytes) | <offset> (8 bytes) | <order> (1 byte) | <mode> (1 byte) | <nsubmodes> (8 bytes) | <submodes> (nsubmodes*1 bytes) | <flags> (4 bytes) |

which translates to the following ArrayBuffer layout:

ArrayBuffer[
    <endianness>[int8],
    <dtype>[int16],
    <ndims>[int64],
    <shape>[ndims*int64],
    <strides>[ndims*int64],
    <offset>[int64],
    <order>[int8],
    <mode>[int8],
    <nsubmodes>[int64],
    <submodes>[nsubmodes*int8],
    <flags>[int32]
]

where strides and offset are in units of bytes.

If the endianness is 1, the byte order is little endian. If the endianness is 0, the byte order is big endian.

Buffer length:

1 + 2 + 8 + (ndims*8) + (ndims*8) + 8 + 1 + 1 + 8 + (nsubmodes*1) + 4 = 33 + (ndims*16) + nsubmodes

For example, consider a three-dimensional ndarray with one subscript index mode (submode):

33 + (3*16) + 1 = 82 bytes

Examples

var IS_LITTLE_ENDIAN = require( '@stdlib/assert-is-little-endian' );
var array = require( '@stdlib/ndarray-array' );
var Uint8Array = require( '@stdlib/array-uint8' );
var fromInt64Bytes = require( '@stdlib/number-float64-base-from-int64-bytes' );
var serialize = require( '@stdlib/ndarray-base-serialize-meta-data' );

// Create an ndarray:
var x = array( [ [ 1, 2 ], [ 3, 4 ] ] );

// Print various properties:
console.log( 'dtype: %s', x.dtype );
console.log( 'ndims: %d', x.ndims );
console.log( 'shape: [ %s ]', x.shape.join( ', ' ) );
console.log( 'strides: [ %s ]', x.strides.join( ', ' ) );
console.log( 'offset: %d', x.offset );
console.log( 'order: %s', x.order );

// Serialize ndarray meta data to a DataView:
var dv = serialize( x );
// returns <DataView>

// Create a Uint8Array view:
var bytes = new Uint8Array( dv.buffer );

// Extract the data type (enum):
var dtype = dv.getInt16( 1, IS_LITTLE_ENDIAN );
console.log( 'dtype (enum): %d', dtype );

// Extract the number of dimensions:
var ndims = fromInt64Bytes( bytes, 1, 3 );
console.log( 'ndims: %d', ndims );

// Extract the shape:
var shape = [];
var i;
for ( i = 0; i < ndims; i++ ) {
    shape.push( fromInt64Bytes( bytes, 1, 11+(i*8) ) );
}
console.log( 'shape: [ %s ]', shape.join( ', ' ) );

// Extract the strides (in units of bytes):
var strides = [];
for ( i = 0; i < ndims; i++ ) {
    strides.push( fromInt64Bytes( bytes, 1, 11+(ndims*8)+(i*8) ) );
}
console.log( 'strides (bytes): [ %s ]', strides.join( ', ' ) );

// Extract the index offset (in bytes):
var offset = fromInt64Bytes( bytes, 1, 11+(ndims*16) );
console.log( 'offset (bytes): %d', offset );

// Extract the order (enum):
var order = dv.getInt8( 19+(ndims*16) );
console.log( 'order (enum): %d', order );

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.