@thi.ng/transducers
Advanced tools
This project is part of the @thi.ng/umbrella monorepo.
This library provides altogether ~120 transducers, reducers, sequence generators (iterators) and other supporting functions for composing data transformation pipelines.
The overall concept and many of the core functions offered here are directly inspired by the original Clojure implementation by Rich Hickey, though the implementation does heavily differ (also in contrast to some other JS based implementations) and dozens of less common, but generally highly useful operators have been added. See full list below.
Furthermore, most transducers & reducers provided here accept an
optional input iterable, which allows them to be used directly as
transformers instead of having to wrap them in one of the execution
functions (i.e. transduce(), reduce(), iterator(), run(),
step()). If called this way, transducer functions will return a
transforming ES6 iterator (generator) and reducing functions will return
a reduced result of the given input iterable.
There's an ongoing multi-part blog series about internals, use cases & patterns of this package, specifically these 3 parts:
Several previously included internal support functions have been migrated to the @thi.ng/arrays package. You'll need to update your imports if you've been using any of these directly. Note that some of these functions also have changes to their arg order.
Functions using randomness now all support an optional PRNG
implementation of the IRandom interface from the
@thi.ng/random
package.
yarn add @thi.ng/transducers
There're several standalone example projects using this library in the /examples directory.
Almost all functions can be imported selectively, but for development purposes full module re-exports are defined.
// full import (not recommended for browser use)
import * as tx from "@thi.ng/transducers";
// selective / single function imports
import { transduce } from "@thi.ng/transducers";
// compose transducer
xform = tx.comp(
tx.filter((x) => (x & 1) > 0), // odd numbers only
tx.distinct(), // distinct numbers only
tx.map((x) => x * 3) // times 3
);
// collect into array (tx.push)
tx.transduce(xform, tx.push(), [1, 2, 3, 4, 5, 4, 3, 2, 1]);
// [ 3, 9, 15 ]
// re-use same xform, but collect into ES6 Set
tx.transduce(xform, tx.conj(), [1, 2, 3, 4, 5, 4, 3, 2, 1]);
// Set { 3, 9, 15 }
// or apply as transforming iterator
// no reduction, only transformations
[...tx.iterator(xform, [1, 2, 3, 4, 5])]
// [ 3, 9, 15]
// alternatively provide an input iterable and
// use xform as transforming iterator
[...tx.filter((x) => /[A-Z]/.test(x), "Hello World!")]
// ["H", "W"]
// single step execution
// returns undefined if transducer returned no result for this input
// returns array if transducer step produced multiple results
f = tx.step(xform);
f(1) // 3
f(2) // undefined
f(3) // 9
f(4) // undefined
f = tx.step(take)
[...tx.filterFuzzy("ho", ["hello", "hallo", "hey", "heyoka"])]
// ["hello", "hallo", "heyoka"]
[...tx.filterFuzzy("hlo", ["hello", "hallo", "hey", "heyoka"])]
// ["hello", "hallo"]
// works with any array-like values & supports custom key extractors
[...tx.filterFuzzy(
[1, 3],
{ key: (x) => x.tags },
[
{ tags: [1, 2, 3] },
{ tags: [2, 3, 4] },
{ tags: [4, 5, 6] },
{ tags: [1, 3, 6] }
]
)]
// [ { tags: [ 1, 2, 3 ] }, { tags: [ 1, 3, 6 ] } ]
// use the `frequencies` reducer to create
// a map counting occurrence of each value
tx.transduce(tx.map((x) => x.toUpperCase()), tx.frequencies(), "hello world");
// Map { 'H' => 1, 'E' => 1, 'L' => 3, 'O' => 2, ' ' => 1, 'W' => 1, 'R' => 1, 'D' => 1 }
// reduction only (no transform)
tx.reduce(tx.frequencies(), [1, 1, 1, 2, 3, 4, 4]);
// Map { 1 => 3, 2 => 1, 3 => 1, 4 => 2 }
// direct reduction if input is given
tx.frequencies([1, 1, 1, 2, 3, 4, 4]);
// Map { 1 => 3, 2 => 1, 3 => 1, 4 => 2 }
// with optional key function, here to bin by word length
tx.frequencies(
(x) => x.length,
"my camel is collapsing and needs some water".split(" ")
);
// Map { 2 => 2, 5 => 3, 10 => 1, 3 => 1, 4 => 1 }
// actual grouping (here: by word length)
tx.groupByMap(
{ key: (x) => x.length },
"my camel is collapsing and needs some water".split(" ")
);
// Map {
// 2 => [ 'my', 'is' ],
// 3 => [ 'and' ],
// 4 => [ 'some' ],
// 5 => [ 'camel', 'needs', 'water' ],
// 10 => [ 'collapsing' ]
// }
// extract only items for given page id & page length
[...tx.page(0, 5, tx.range(12))]
// [ 0, 1, 2, 3, 4 ]
// when composing with other transducers
// it's most efficient to place `page()` early on in the chain
// that way only the page items will be further processed
[...tx.iterator(tx.comp(tx.page(1, 5), tx.map(x => x * 10)), tx.range(12))]
// [ 50, 60, 70, 80, 90 ]
// use `padLast()` to fill up missing values
[...tx.iterator(tx.comp(tx.page(2, 5), tx.padLast(5, "n/a")), tx.range(12))]
// [ 10, 11, 'n/a', 'n/a', 'n/a' ]
// no values produced for invalid pages
[...tx.page(3, 5, tx.range(12))]
// []
multiplex and multiplexObj can be used to transform values in
parallel using the provided transducers (which can be composed as usual)
and results in a tuple or keyed object.
tx.transduce(
tx.multiplex(
tx.map((x) => x.charAt(0)),
tx.map((x) => x.toUpperCase()),
tx.map((x) => x.length)
),
tx.push(),
["Alice", "Bob", "Charlie"]
);
// [ [ "A", "ALICE", 5 ], [ "B", "BOB", 3 ], [ "C", "CHARLIE", 7 ] ]
tx.transduce(
tx.multiplexObj({
initial: tx.map((x) => x.charAt(0)),
name: tx.map((x) => x.toUpperCase()),
len: tx.map((x) => x.length)
}),
tx.push(),
["Alice", "Bob", "Charlie"]
);
// [ { len: 5, name: 'ALICE', initial: 'A' },
// { len: 3, name: 'BOB', initial: 'B' },
// { len: 7, name: 'CHARLIE', initial: 'C' } ]
// use nested reduce to compute window averages
tx.transduce(
tx.comp(
tx.partition(5, 1),
tx.map(x => tx.reduce(tx.mean(), x))
),
tx.push(),
[1, 2, 3, 3, 4, 5, 5, 6, 7, 8, 8, 9, 10]
)
// [ 2.6, 3.4, 4, 4.6, 5.4, 6.2, 6.8, 7.6, 8.4 ]
// this combined transducer is also directly
// available as: `tx.movingAverage(n)`
[...tx.movingAverage(5, [1, 2, 3, 3, 4, 5, 5, 6, 7, 8, 8, 9, 10])]
// [ 2.6, 3.4, 4, 4.6, 5.4, 6.2, 6.8, 7.6, 8.4 ]
// function to test
fn = () => {
let x;
for (i = 0; i < 1e6; i++) {
x = Math.cos(i);
}
return x;
};
// compute the mean of 100 runs
tx.transduce(tx.benchmark(), tx.mean(), tx.repeatedly(fn, 100));
// 1.93 (milliseconds)
// alternatively, use tx.sideEffect() for any side fx
tx.transduce(
tx.comp(
tx.trace("orig"),
tx.map((x) => x + 1),
tx.trace("mapped"),
tx.filter((x) => (x & 1) > 0)
),
tx.push(),
[1, 2, 3, 4]
);
// orig 1
// mapped 2
// orig 2
// mapped 3
// orig 3
// mapped 4
// orig 4
// mapped 5
// [ 3, 5 ]
The struct transducer is simply a composition of: partitionOf -> partition -> rename -> mapKeys. See code
here.
// Higher-order transducer to convert linear input into structured objects
// using given field specs and ordering. A single field spec is an array of
// 2 or 3 items: `[name, size, transform?]`. If `transform` is given, it will
// be used to produce the final value for this field. In the example below,
// it is used to unwrap the ID field values, e.g. from `[0] => 0`
[
...tx.struct(
[["id", 1, (id) => id[0]], ["pos", 2], ["vel", 2], ["color", 4]],
[0, 100, 200, -1, 0, 1, 0.5, 0, 1, 1, 0, 0, 5, 4, 0, 0, 1, 1]
)
];
// [ { color: [ 1, 0.5, 0, 1 ],
// vel: [ -1, 0 ],
// pos: [ 100, 200 ],
// id: 0 },
// { color: [ 0, 0, 1, 1 ],
// vel: [ 5, 4 ],
// pos: [ 0, 0 ],
// id: 1 } ]
tx.transduce(
tx.comp(
// split into rows
tx.mapcat((x) => x.split("\n")),
// split each row
tx.map((x) => x.split(",")),
// convert each row into object, rename array indices
tx.rename({ id: 0, name: 1, alias: 2, num: "length" })
),
tx.push(),
["100,typescript\n101,clojure,clj\n110,rust,rs"]
);
// [ { num: 2, name: 'typescript', id: '100' },
// { num: 3, alias: 'clj', name: 'clojure', id: '101' },
// { num: 3, alias: 'rs', name: 'rust', id: '110' } ]
// result is realized after max. 7 values, irrespective of nesting
tx.transduce(tx.comp(tx.flatten(), tx.take(7)), tx.push(), [
1,
[2, [3, 4, [5, 6, [7, 8], 9, [10]]]]
]);
// [1, 2, 3, 4, 5, 6, 7]
// this transducer uses 2 scans (a scan = inner reducer per item)
// 1) counts incoming values
// 2) forms an array of the current counter value `x` & repeated `x` times
// 3) emits results as series of reductions in the outer array produced
// by the main reducer
// IMPORTANT: since arrays are mutable we use `pushCopy` as the inner reducer
// instead of `push` (the outer reducer)
xform = tx.comp(
tx.scan(tx.count()),
tx.map(x => [...tx.repeat(x,x)]),
tx.scan(tx.pushCopy())
)
[...tx.iterator(xform, [1, 1, 1, 1])]
// [ [ [ 1 ] ],
// [ [ 1 ], [ 2, 2 ] ],
// [ [ 1 ], [ 2, 2 ], [ 3, 3, 3 ] ],
// [ [ 1 ], [ 2, 2 ], [ 3, 3, 3 ], [ 4, 4, 4, 4 ] ] ]
// more simple & similar to previous, but without the 2nd xform step
tx.transduce(tx.comp(tx.scan(tx.count()), tx.scan(tx.pushCopy())), tx.push(), [1,1,1,1])
// [ [ 1 ], [ 1, 2 ], [ 1, 2, 3 ], [ 1, 2, 3, 4 ] ]
[...tx.take(10, tx.choices("abcd", [1, 0.5, 0.25, 0.125]))];
// [ 'a', 'a', 'b', 'a', 'a', 'b', 'a', 'c', 'd', 'b' ]
tx.transduce(
tx.take(1000),
tx.frequencies(),
tx.choices("abcd", [1, 0.5, 0.25, 0.125])
);
// Map { 'c' => 132, 'a' => 545, 'b' => 251, 'd' => 72 }
See
interpolate()
docs for details.
[
...interpolate(
10,
0,
100,
(a, b) => [a, b],
([a, b], t) => Math.floor(a + (b - a) * t),
[20, 100],
[50, 200],
[80, 0]
)
];
// [ 100, 100, 100, 133, 166, 200, 133, 66, 0, 0, 0 ]
Documentation is slowly forthcoming in the form of doc comments (incl. code examples) for a growing number the functions listed below. Please see source code for now.
Apart from type aliases, the only real types defined are:
Reducers are the core building blocks of transducers. Unlike other
implementations using OOP approaches, a Reducer in this lib is a
simple 3-element array of functions, each addressing a separate
processing step.
Since v0.6.0 the bundled reducers are all wrapped in functions to provide a uniform API (and some of them can be preconfigured and/or are stateful closures). However, it's fine to define stateless reducers as constant arrays.
interface Reducer<A, B> extends Array<any> {
/**
* Initialization, e.g. to provide a suitable accumulator value,
* only called when no initial accumulator has been provided by user.
*/
[0]: () => A;
/**
* Completion. When called usually just returns `acc`, but stateful
* transformers should flush/apply their outstanding results.
*/
[1]: (acc: A) => A;
/**
* Reduction step. Combines new input with accumulator.
* If reduction should terminate early, wrap result via `reduced()`
*/
[2]: (acc: A, x: B) => A | Reduced<A>;
}
// A concrete example:
const push: Reducer<any[], any> = [
// init
() => [],
// completion (nothing to do in this case)
(acc) => acc,
// step
(acc, x) => (acc.push(x), acc)
];
partition, partitionBy, streamSort, streamShuffle are (examples
of) transducers making use of their 1-arity completing function.
class Reduced<T> implements IDeref<T> {
protected value: T;
constructor(val: T);
deref(): T;
}
Simple type wrapper to identify early termination of a reducer. Does not
modify wrapped value by injecting magic properties. Instances can be
created via reduced(x) and handled via these helper functions:
reduced(x: any): anyisReduced(x: any): booleanensureReduced(x: any): Reduced<any>unreduced(x: any): anyBy default reduce() consumes inputs via the standard ES6 Iterable
interface, i.e. using a for..of.. loop. Array-like inputs are consumed
via a traditional for-loop and custom optimized iterations can be
provided via implementations of the IReducible interface in the source
collection type. Examples can be found here:
Note: The IReducible interface is only used by reduce(),
transduce() and run().
From Rich Hickey's original definition:
A transducer is a transformation from one reducing function to another
As shown in the examples above, transducers can be dynamically composed
(using comp()) to form arbitrary data transformation pipelines without
causing large overheads for intermediate collections.
type Transducer<A, B> = (rfn: Reducer<any, B>) => Reducer<any, A>;
// concrete example of stateless transducer (expanded for clarity)
function map<A, B>(fn: (x: A) => B): Transducer<A, B> {
return (rfn: Reducer<any, B>) => {
return [
() => rfn[0](),
(acc) => rfn[1](acc),
(acc, x: A) => rfn[2](acc, fn(x))
];
};
}
// stateful transducer
// removes successive value repetitions
function dedupe<T>(): Transducer<T, T> {
return (rfn: Reducer<any, T>) => {
// state initialization
let prev = {};
return [
() => rfn[0](),
(acc) => rfn[1](acc),
(acc, x) => {
acc = prev === x ? acc : rfn[2](acc, x);
prev = x;
return acc;
}
];
};
}
comp(f1, f2, ...)Returns new transducer composed from given transducers. Data flow is from left to right. Offers fast paths for up to 10 args. If more are given, composition is done dynamically via for loop.
compR(rfn: Reducer<any, any>, fn: (acc, x) => any): Reducer<any, any>Helper function to compose reducers.
iterator<A, B>(tx: Transducer<A, B>, xs: Iterable<A>): IterableIterator<B>Similar to transduce(), but emits results as ES6 iterator (and hence
doesn't use a reduction function).
reduce<A, B>(rfn: Reducer<A, B>, acc: A, xs: Iterable<B>): AReduces xs using given reducer and optional initial
accumulator/result. If xs implements the IReducible interface,
delegates to that implementation. Likewise, uses a fast route if xs is
an ArrayLike type.
transduce<A, B, C>(tx: Transducer<A, B>, rfn: Reducer<C, B>, acc: C, xs: Iterable<A>): CTransforms iterable using given transducer and combines results with given reducer and optional initial accumulator/result.
run<A, B>(tx: Transducer<A, B>, fx: (x: B) => void, xs: Iterable<A>)Transforms iterable with given transducer and optional side effect
without any reduction step. If fx is given it will be called with
every value produced by the transducer. If fx is not given, the
transducer is assumed to include at least one sideEffect() step
itself. Returns nothing.
All of the following functions can be used and composed as transducers. With a few exceptions, most also accept an input iterable and then directly yield a transforming iterator, e.g.
// as transducer
tx.transduce(tx.map((x) => x*10), tx.push(), tx.range(4))
// [ 0, 10, 20, 30 ]
// as transforming iterator
[...tx.map((x) => x*10, tx.range(4))]
// [ 0, 10, 20, 30 ]
zip)As with transducer functions, reducer functions can also given an
optional input iterable. If done so, the function will consume the input
and return a reduced result (as if it would be called via reduce()).
© 2016-2018 Karsten Schmidt // Apache Software License 2.0
FAQs
Collection of ~170 lightweight, composable transducers, reducers, generators, iterators for functional data transformations
We found that @thi.ng/transducers demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 0 open source maintainers collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Research
Security News
Socket researchers uncover a malicious npm package posing as a tool for detecting vulnerabilities in Etherium smart contracts.
Security News
Research
A supply chain attack on Rspack's npm packages injected cryptomining malware, potentially impacting thousands of developers.
Research
Security News
Socket researchers discovered a malware campaign on npm delivering the Skuld infostealer via typosquatted packages, exposing sensitive data.