Awesome

unionizor

Superfast union supporting Arrays and Iterables containing primitives or objects. In the age of big data, you need it.

Performance

Unionizer is extraordinarily fast. Multiple test runs show unionizer is 60-100x faster when accessing the first element of a union, 30-50x faster when accessing the central point of a union, and 10-15% slower than lodash when accessing the last item of union but still 10-15% faster than setUnion. If you are running a data processing pipeline, time to first response is critical. Secondary access of the same item will be sub millisecond since unionizer caches values in an array.

The results below are for the union of six 100,000 element arrays containing random numbers.

The function setUnion is defined as:

function setUnion(...iterables) {
	return [...iterables.reduce((set,arg) => new Set([...set,...arg]))]
}

All approaches have variable speeds due to garbage collection.

Below are results from a sample run:

lodash create 85827 from 6 x 100000: 68.936ms
lodash access 0: 69.903ms
lodash access 42913.5: 70.358ms
lodash access 85826: 70.735ms
lodash re-access 85826: 0.121ms
lodash total: 71.626ms

setUnion create 85827 from 6 x 100000: 102.689ms
setUnion access 0: 103.528ms
setUnion access 42913.5: 103.933ms
setUnion access 85826: 104.287ms
setUnion re-access 85826: 0.107ms
setUnion total: 105.124ms

unionizer create 85827 from 6 x 100000: 0.047ms
unionizer access 0: 0.544ms
unionizer access 42913.5: 0.886ms
unionizer access 85826: 76.758ms
unionizer re-access 85826: 0.335ms
unionizer total: 78.063ms

If you want similar performance for intersection, Cartesian product, or memoizing also see:

• https://github.com/anywhichway/intersector
• https://github.com/anywhichway/cxproduct
• https://github.com/anywhichway/nano-memoize

For a complete high performance solution to set operations for Arrays and Sets with a standardized API, plus the addition of the standard map/reduce/find operations to Set see:

• https://github.com/anywhichway/array-set-ops

Installing

npm install unionizor

Using

iterable <- unionizor(booleanOrUniqueKeyProperty=false) is a function that returns another function configured to do array unions and return arrays. It takes one optional argument booleanOrUniqueKeyProperty that defaults to `false'. If booleanOrUniqueKeyProperty=false the intersection is optimized for direct value comparison of primitives or objects (not deepequals). If typeof(booleanOrUniqueKeyProperty)==="string" it is assumed to be a unique key on all objects in the arrays, primitives will still be handled.

import unionizor from "unionizer";
const union = unionizor(),
	keyedObjectUnion = unionizor("o");
union([1,2,3],new Set([3,2])); // will return [1,2,3]
var o1 = {o:1},
    o2 = {o:2},
    o3 = {o:3};
keyedObjectUnion([o1,o2,o3],[o3,o2]); // will return [o1,o2,o3];

Note: Accessing the length property on a unionizer union forces full resolution of the union and will slow down your processing. Use a for(...of...) loop or

Architecture

By definition a union contains one of every item that occurs in every group of items added to the union. As a result, it is not necessary to compute the entire union prior to returning the first item. It is only necessary to ensure that duplicates are not returned. Hence, an iterable is implemented that keeps track of items returned and skips them. Returned items are stored in an array so that subsequent access is faster.

Custom versions of forEach, map, and reduce are implemented to avoid accessing length.

Test Coverage

15 passing (322ms)

----------|---------|----------|---------|---------|-------------------

Updates (reverse chronological order)

2022-02-22 v2.0.2 - Updated documentation.

2022-02-15 v2.0.1 - Updated documentation.

2022-02-15 v2.0.0 - Updated to use iterable protocol. More than 10x faster overall.

2017-01-15 v1.0.3 - Added Codacy quality checks and badges. Replace var with let. Addressed a package loading issues that impacted testing.

2017-01-02 v1.0.2 - Previous versions would treat string equivalents in primitive arrays the same, e.g. 2 would equal "2". Also, improved performance and added the option to pass in an object key.

2017-01-01 v1.0.1 - Externalized benchmark.

2016-12-27 v1.0.0 - Initial public release.

License

MIT - see LICENSE file