Collection Functions (Arrays or Objects)

each_.each(list, iterator, [context]) Alias: forEach
Iterates over a list of elements, yielding each in turn to an iterator function. The iterator is bound to the context object, if one is passed. Each invocation of iterator is called with three arguments: (element, index, list). If list is a JavaScript object, iterator's arguments will be (value, key, list). Delegates to the native forEach function if it exists.

_.each([1, 2, 3], function(num){ alert(num); });
=> alerts each number in turn...
_.each({one : 1, two : 2, three : 3}, function(num, key){ alert(num); });
=> alerts each number in turn...

map_.map(list, iterator, [context]) Alias: collect
Produces a new array of values by mapping each value in list through a transformation function (iterator). If the native map method exists, it will be used instead. If list is a JavaScript object, iterator's arguments will be (value, key, list).

_.map([1, 2, 3], function(num){ return num * 3; });
=> [3, 6, 9]
_.map({one : 1, two : 2, three : 3}, function(num, key){ return num * 3; });
=> [3, 6, 9]

reduce_.reduce(list, iterator, memo, [context]) Aliases: inject, foldl
Also known as inject and foldl, reduce boils down a list of values into a single value. Memo is the initial state of the reduction, and each successive step of it should be returned by iterator. The iterator is passed four arguments: the memo, then the value and index (or key) of the iteration, and finally a reference to the entire list.

var sum = _.reduce([1, 2, 3], function(memo, num){ return memo + num; }, 0);
=> 6

reduceRight_.reduceRight(list, iterator, memo, [context]) Alias: foldr
The right-associative version of reduce. Delegates to the JavaScript 1.8 version of reduceRight, if it exists. Foldr is not as useful in JavaScript as it would be in a language with lazy evaluation.

var list = [[0, 1], [2, 3], [4, 5]];
var flat = _.reduceRight(list, function(a, b) { return a.concat(b); }, []);
=> [4, 5, 2, 3, 0, 1]

find_.find(list, iterator, [context]) Alias: detect
Looks through each value in the list, returning the first one that passes a truth test (iterator). The function returns as soon as it finds an acceptable element, and doesn't traverse the entire list.

var even = _.find([1, 2, 3, 4, 5, 6], function(num){ return num % 2 == 0; });
=> 2

filter_.filter(list, iterator, [context]) Alias: select
Looks through each value in the list, returning an array of all the values that pass a truth test (iterator). Delegates to the native filter method, if it exists.

var evens = _.filter([1, 2, 3, 4, 5, 6], function(num){ return num % 2 == 0; });
=> [2, 4, 6]

where_.where(list, properties)
Looks through each value in the list, returning an array of all the values that contain all of the key-value pairs listed in properties.

_.where(listOfPlays, {author: "Shakespeare", year: 1611});
=> [{title: "Cymbeline", author: "Shakespeare", year: 1611},
    {title: "The Tempest", author: "Shakespeare", year: 1611}]

reject_.reject(list, iterator, [context])
Returns the values in list without the elements that the truth test (iterator) passes. The opposite of filter.

var odds = _.reject([1, 2, 3, 4, 5, 6], function(num){ return num % 2 == 0; });
=> [1, 3, 5]

all_.all(list, iterator, [context]) Alias: every
Returns true if all of the values in the list pass the iterator truth test. Delegates to the native method every, if present.

_.all([true, 1, null, 'yes'], _.identity);
=> false

any_.any(list, [iterator], [context]) Alias: some
Returns true if any of the values in the list pass the iterator truth test. Short-circuits and stops traversing the list if a true element is found. Delegates to the native method some, if present.

_.any([null, 0, 'yes', false]);
=> true

contains_.contains(list, value) Alias: include
Returns true if the value is present in the list. Uses indexOf internally, if list is an Array.

_.contains([1, 2, 3], 3);
=> true

invoke_.invoke(list, methodName, [*arguments])
Calls the method named by methodName on each value in the list. Any extra arguments passed to invoke will be forwarded on to the method invocation.

_.invoke([[5, 1, 7], [3, 2, 1]], 'sort');
=> [[1, 5, 7], [1, 2, 3]]

pluck_.pluck(list, propertyName)
A convenient version of what is perhaps the most common use-case for map: extracting a list of property values.

var stooges = [{name : 'moe', age : 40}, {name : 'larry', age : 50}, {name : 'curly', age : 60}];
_.pluck(stooges, 'name');
=> ["moe", "larry", "curly"]

max_.max(list, [iterator], [context])
Returns the maximum value in list. If iterator is passed, it will be used on each value to generate the criterion by which the value is ranked.

var stooges = [{name : 'moe', age : 40}, {name : 'larry', age : 50}, {name : 'curly', age : 60}];
_.max(stooges, function(stooge){ return stooge.age; });
=> {name : 'curly', age : 60};

min_.min(list, [iterator], [context])
Returns the minimum value in list. If iterator is passed, it will be used on each value to generate the criterion by which the value is ranked.

var numbers = [10, 5, 100, 2, 1000];
_.min(numbers);
=> 2

sortBy_.sortBy(list, iterator, [context])
Returns a sorted copy of list, ranked in ascending order by the results of running each value through iterator. Iterator may also be the string name of the property to sort by (eg. length).

_.sortBy([1, 2, 3, 4, 5, 6], function(num){ return Math.sin(num); });
=> [5, 4, 6, 3, 1, 2]

groupBy_.groupBy(list, iterator)
Splits a collection into sets, grouped by the result of running each value through iterator. If iterator is a string instead of a function, groups by the property named by iterator on each of the values.

_.groupBy([1.3, 2.1, 2.4], function(num){ return Math.floor(num); });
=> {1: [1.3], 2: [2.1, 2.4]}

_.groupBy(['one', 'two', 'three'], 'length');
=> {3: ["one", "two"], 5: ["three"]}

countBy_.countBy(list, iterator)
Sorts a list into groups and returns a count for the number of objects in each group. Similar to groupBy, but instead of returning a list of values, returns a count for the number of values in that group.

_.countBy([1, 2, 3, 4, 5], function(num) { 
  return num % 2 == 0 ? 'even' : 'odd'; 
});
=> {odd: 3, even: 2}

shuffle_.shuffle(list)
Returns a shuffled copy of the list, using a version of the Fisher-Yates shuffle.

_.shuffle([1, 2, 3, 4, 5, 6]);
=> [4, 1, 6, 3, 5, 2]

toArray_.toArray(list)
Converts the list (anything that can be iterated over), into a real Array. Useful for transmuting the arguments object.

(function(){ return _.toArray(arguments).slice(1); })(1, 2, 3, 4);
=> [2, 3, 4]

size_.size(list)
Return the number of values in the list.

_.size({one : 1, two : 2, three : 3});
=> 3

Array Functions

Note: All array functions will also work on the arguments object. However, Underscore functions are not designed to work on "sparse" arrays.

first_.first(array, [n]) Alias: head, take
Returns the first element of an array. Passing n will return the first n elements of the array.

_.first([5, 4, 3, 2, 1]);
=> 5

initial_.initial(array, [n])
Returns everything but the last entry of the array. Especially useful on the arguments object. Pass n to exclude the last n elements from the result.

_.initial([5, 4, 3, 2, 1]);
=> [5, 4, 3, 2]

last_.last(array, [n])
Returns the last element of an array. Passing n will return the last n elements of the array.

_.last([5, 4, 3, 2, 1]);
=> 1

rest_.rest(array, [index]) Alias: tail, drop
Returns the rest of the elements in an array. Pass an index to return the values of the array from that index onward.

_.rest([5, 4, 3, 2, 1]);
=> [4, 3, 2, 1]

compact_.compact(array)
Returns a copy of the array with all falsy values removed. In JavaScript, false, null, 0, "", undefined and NaN are all falsy.

_.compact([0, 1, false, 2, '', 3]);
=> [1, 2, 3]

flatten_.flatten(array, [shallow])
Flattens a nested array (the nesting can be to any depth). If you pass shallow, the array will only be flattened a single level.

_.flatten([1, [2], [3, [[4]]]]);
=> [1, 2, 3, 4];

_.flatten([1, [2], [3, [[4]]]], true);
=> [1, 2, 3, [[4]]];

without_.without(array, [*values])
Returns a copy of the array with all instances of the values removed.

_.without([1, 2, 1, 0, 3, 1, 4], 0, 1);
=> [2, 3, 4]

union_.union(*arrays)
Computes the union of the passed-in arrays: the list of unique items, in order, that are present in one or more of the arrays.

_.union([1, 2, 3], [101, 2, 1, 10], [2, 1]);
=> [1, 2, 3, 101, 10]

intersection_.intersection(*arrays)
Computes the list of values that are the intersection of all the arrays. Each value in the result is present in each of the arrays.

_.intersection([1, 2, 3], [101, 2, 1, 10], [2, 1]);
=> [1, 2]

difference_.difference(array, *others)
Similar to without, but returns the values from array that are not present in the other arrays.

_.difference([1, 2, 3, 4, 5], [5, 2, 10]);
=> [1, 3, 4]

uniq_.uniq(array, [isSorted], [iterator]) Alias: unique
Produces a duplicate-free version of the array, using === to test object equality. If you know in advance that the array is sorted, passing true for isSorted will run a much faster algorithm. If you want to compute unique items based on a transformation, pass an iterator function.

_.uniq([1, 2, 1, 3, 1, 4]);
=> [1, 2, 3, 4]

zip_.zip(*arrays)
Merges together the values of each of the arrays with the values at the corresponding position. Useful when you have separate data sources that are coordinated through matching array indexes. If you're working with a matrix of nested arrays, zip.apply can transpose the matrix in a similar fashion.

_.zip(['moe', 'larry', 'curly'], [30, 40, 50], [true, false, false]);
=> [["moe", 30, true], ["larry", 40, false], ["curly", 50, false]]

object_.object(list, [values])
Converts arrays into objects. Pass either a single list of [key, value] pairs, or a list of keys, and a list of values.

_.object(['moe', 'larry', 'curly'], [30, 40, 50]);
=> {moe: 30, larry: 40, curly: 50}

_.object([['moe', 30], ['larry', 40], ['curly', 50]]);
=> {moe: 30, larry: 40, curly: 50}

indexOf_.indexOf(array, value, [isSorted])
Returns the index at which value can be found in the array, or -1 if value is not present in the array. Uses the native indexOf function unless it's missing. If you're working with a large array, and you know that the array is already sorted, pass true for isSorted to use a faster binary search ... or, pass a number as the third argument in order to look for the first matching value in the array after the given index.

_.indexOf([1, 2, 3], 2);
=> 1

lastIndexOf_.lastIndexOf(array, value, [fromIndex])
Returns the index of the last occurrence of value in the array, or -1 if value is not present. Uses the native lastIndexOf function if possible. Pass fromIndex to start your search at a given index.

_.lastIndexOf([1, 2, 3, 1, 2, 3], 2);
=> 4

sortedIndex_.sortedIndex(list, value, [iterator])
Uses a binary search to determine the index at which the value should be inserted into the list in order to maintain the list's sorted order. If an iterator is passed, it will be used to compute the sort ranking of each value, including the value you pass.

_.sortedIndex([10, 20, 30, 40, 50], 35);
=> 3

range_.range([start], stop, [step])
A function to create flexibly-numbered lists of integers, handy for each and map loops. start, if omitted, defaults to 0; step defaults to 1. Returns a list of integers from start to stop, incremented (or decremented) by step, exclusive.

_.range(10);
=> [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
_.range(1, 11);
=> [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
_.range(0, 30, 5);
=> [0, 5, 10, 15, 20, 25]
_.range(0, -10, -1);
=> [0, -1, -2, -3, -4, -5, -6, -7, -8, -9]
_.range(0);
=> []

Function (uh, ahem) Functions

bind_.bind(function, object, [*arguments])
Bind a function to an object, meaning that whenever the function is called, the value of this will be the object. Optionally, bind arguments to the function to pre-fill them, also known as partial application.

var func = function(greeting){ return greeting + ': ' + this.name };
func = _.bind(func, {name : 'moe'}, 'hi');
func();
=> 'hi: moe'

bindAll_.bindAll(object, [*methodNames])
Binds a number of methods on the object, specified by methodNames, to be run in the context of that object whenever they are invoked. Very handy for binding functions that are going to be used as event handlers, which would otherwise be invoked with a fairly useless this. If no methodNames are provided, all of the object's function properties will be bound to it.

var buttonView = {
  label   : 'underscore',
  onClick : function(){ alert('clicked: ' + this.label); },
  onHover : function(){ console.log('hovering: ' + this.label); }
};
_.bindAll(buttonView);
jQuery('#underscore_button').bind('click', buttonView.onClick);
=> When the button is clicked, this.label will have the correct value...

memoize_.memoize(function, [hashFunction])
Memoizes a given function by caching the computed result. Useful for speeding up slow-running computations. If passed an optional hashFunction, it will be used to compute the hash key for storing the result, based on the arguments to the original function. The default hashFunction just uses the first argument to the memoized function as the key.

var fibonacci = _.memoize(function(n) {
  return n < 2 ? n : fibonacci(n - 1) + fibonacci(n - 2);
});

delay_.delay(function, wait, [*arguments])
Much like setTimeout, invokes function after wait milliseconds. If you pass the optional arguments, they will be forwarded on to the function when it is invoked.

var log = _.bind(console.log, console);
_.delay(log, 1000, 'logged later');
=> 'logged later' // Appears after one second.

defer_.defer(function, [*arguments])
Defers invoking the function until the current call stack has cleared, similar to using setTimeout with a delay of 0. Useful for performing expensive computations or HTML rendering in chunks without blocking the UI thread from updating. If you pass the optional arguments, they will be forwarded on to the function when it is invoked.

_.defer(function(){ alert('deferred'); });
// Returns from the function before the alert runs.

throttle_.throttle(function, wait)
Creates and returns a new, throttled version of the passed function, that, when invoked repeatedly, will only actually call the original function at most once per every wait milliseconds. Useful for rate-limiting events that occur faster than you can keep up with.

var throttled = _.throttle(updatePosition, 100);
$(window).scroll(throttled);

debounce_.debounce(function, wait, [immediate])
Creates and returns a new debounced version of the passed function that will postpone its execution until after wait milliseconds have elapsed since the last time it was invoked. Useful for implementing behavior that should only happen after the input has stopped arriving. For example: rendering a preview of a Markdown comment, recalculating a layout after the window has stopped being resized, and so on.

Pass true for the immediate parameter to cause debounce to trigger the function on the leading instead of the trailing edge of the wait interval. Useful in circumstances like preventing accidental double-clicks on a "submit" button from firing a second time.

var lazyLayout = _.debounce(calculateLayout, 300);
$(window).resize(lazyLayout);

once_.once(function)
Creates a version of the function that can only be called one time. Repeated calls to the modified function will have no effect, returning the value from the original call. Useful for initialization functions, instead of having to set a boolean flag and then check it later.

var initialize = _.once(createApplication);
initialize();
initialize();
// Application is only created once.

after_.after(count, function)
Creates a version of the function that will only be run after first being called count times. Useful for grouping asynchronous responses, where you want to be sure that all the async calls have finished, before proceeding.

var renderNotes = _.after(notes.length, render);
_.each(notes, function(note) {
  note.asyncSave({success: renderNotes});
});
// renderNotes is run once, after all notes have saved.

wrap_.wrap(function, wrapper)
Wraps the first function inside of the wrapper function, passing it as the first argument. This allows the wrapper to execute code before and after the function runs, adjust the arguments, and execute it conditionally.

var hello = function(name) { return "hello: " + name; };
hello = _.wrap(hello, function(func) {
  return "before, " + func("moe") + ", after";
});
hello();
=> 'before, hello: moe, after'

compose_.compose(*functions)
Returns the composition of a list of functions, where each function consumes the return value of the function that follows. In math terms, composing the functions f(), g(), and h() produces f(g(h())).

var greet    = function(name){ return "hi: " + name; };
var exclaim  = function(statement){ return statement + "!"; };
var welcome = _.compose(exclaim, greet);
welcome('moe');
=> 'hi: moe!'

Object Functions

keys_.keys(object)
Retrieve all the names of the object's properties.

_.keys({one : 1, two : 2, three : 3});
=> ["one", "two", "three"]

values_.values(object)
Return all of the values of the object's properties.

_.values({one : 1, two : 2, three : 3});
=> [1, 2, 3]

pairs_.pairs(object)
Convert an object into a list of [key, value] pairs.

_.pairs({one: 1, two: 2, three: 3});
=> [["one", 1], ["two", 2], ["three", 3]]

invert_.invert(object)
Returns a copy of the object where the keys have become the values and the values the keys. For this to work, all of your object's values should be unique and string serializable.

_.invert({Moe: "Moses", Larry: "Louis", Curly: "Jerome"});
=> {Moses: "Moe", Louis: "Larry", Jerome: "Curly"};

functions_.functions(object) Alias: methods
Returns a sorted list of the names of every method in an object — that is to say, the name of every function property of the object.

_.functions(_);
=> ["all", "any", "bind", "bindAll", "clone", "compact", "compose" ...

extend_.extend(destination, *sources)
Copy all of the properties in the source objects over to the destination object, and return the destination object. It's in-order, so the last source will override properties of the same name in previous arguments.

_.extend({name : 'moe'}, {age : 50});
=> {name : 'moe', age : 50}

pick_.pick(object, *keys)
Return a copy of the object, filtered to only have values for the whitelisted keys (or array of valid keys).

_.pick({name : 'moe', age: 50, userid : 'moe1'}, 'name', 'age');
=> {name : 'moe', age : 50}

omit_.omit(object, *keys)
Return a copy of the object, filtered to omit the blacklisted keys (or array of keys).

_.omit({name : 'moe', age : 50, userid : 'moe1'}, 'userid');
=> {name : 'moe', age : 50}

defaults_.defaults(object, *defaults)
Fill in null and undefined properties in object with values from the defaults objects, and return the object. As soon as the property is filled, further defaults will have no effect.

var iceCream = {flavor : "chocolate"};
_.defaults(iceCream, {flavor : "vanilla", sprinkles : "lots"});
=> {flavor : "chocolate", sprinkles : "lots"}

clone_.clone(object)
Create a shallow-copied clone of the object. Any nested objects or arrays will be copied by reference, not duplicated.

_.clone({name : 'moe'});
=> {name : 'moe'};

tap_.tap(object, interceptor)
Invokes interceptor with the object, and then returns object. The primary purpose of this method is to "tap into" a method chain, in order to perform operations on intermediate results within the chain.

_.chain([1,2,3,200])
  .filter(function(num) { return num % 2 == 0; })
  .tap(alert)
  .map(function(num) { return num * num })
  .value();
=> // [2, 200] (alerted)
=> [4, 40000]

has_.has(object, key)
Does the object contain the given key? Identical to object.hasOwnProperty(key), but uses a safe reference to the hasOwnProperty function, in case it's been overridden accidentally.

_.has({a: 1, b: 2, c: 3}, "b");
=> true

isEqual_.isEqual(object, other)
Performs an optimized deep comparison between the two objects, to determine if they should be considered equal.

var moe   = {name : 'moe', luckyNumbers : [13, 27, 34]};
var clone = {name : 'moe', luckyNumbers : [13, 27, 34]};
moe == clone;
=> false
_.isEqual(moe, clone);
=> true

isEmpty_.isEmpty(object)
Returns true if object contains no values.

_.isEmpty([1, 2, 3]);
=> false
_.isEmpty({});
=> true

isElement_.isElement(object)
Returns true if object is a DOM element.

_.isElement(jQuery('body')[0]);
=> true

isArray_.isArray(object)
Returns true if object is an Array.

(function(){ return _.isArray(arguments); })();
=> false
_.isArray([1,2,3]);
=> true

isObject_.isObject(value)
Returns true if value is an Object. Note that JavaScript arrays and functions are objects, while (normal) strings and numbers are not.

_.isObject({});
=> true
_.isObject(1);
=> false

isArguments_.isArguments(object)
Returns true if object is an Arguments object.

(function(){ return _.isArguments(arguments); })(1, 2, 3);
=> true
_.isArguments([1,2,3]);
=> false

isFunction_.isFunction(object)
Returns true if object is a Function.

_.isFunction(alert);
=> true

isString_.isString(object)
Returns true if object is a String.

_.isString("moe");
=> true

isNumber_.isNumber(object)
Returns true if object is a Number (including NaN).

_.isNumber(8.4 * 5);
=> true

isFinite_.isFinite(object)
Returns true if object is a finite Number.

_.isFinite(-101);
=> true

_.isFinite(-Infinity);
=> false

isBoolean_.isBoolean(object)
Returns true if object is either true or false.

_.isBoolean(null);
=> false

isDate_.isDate(object)
Returns true if object is a Date.

_.isDate(new Date());
=> true

isRegExp_.isRegExp(object)
Returns true if object is a RegExp.

_.isRegExp(/moe/);
=> true

isNaN_.isNaN(object)
Returns true if object is NaN.
Note: this is not the same as the native isNaN function, which will also return true if the variable is undefined.

_.isNaN(NaN);
=> true
isNaN(undefined);
=> true
_.isNaN(undefined);
=> false

isNull_.isNull(object)
Returns true if the value of object is null.

_.isNull(null);
=> true
_.isNull(undefined);
=> false

isUndefined_.isUndefined(value)
Returns true if value is undefined.

_.isUndefined(window.missingVariable);
=> true

Utility Functions

noConflict_.noConflict()
Give control of the "_" variable back to its previous owner. Returns a reference to the Underscore object.

var underscore = _.noConflict();

identity_.identity(value)
Returns the same value that is used as the argument. In math: f(x) = x
This function looks useless, but is used throughout Underscore as a default iterator.

var moe = {name : 'moe'};
moe === _.identity(moe);
=> true

times_.times(n, iterator, [context])
Invokes the given iterator function n times. Each invocation of iterator is called with an index argument.

_(3).times(function(n){ genie.grantWishNumber(n); });

random_.random(min, max)
Returns a random integer between min and max, inclusive. If you only pass one argument, it will return a number between 0 and that number.

_.random(0, 100);
=> 42

mixin_.mixin(object)
Allows you to extend Underscore with your own utility functions. Pass a hash of {name: function} definitions to have your functions added to the Underscore object, as well as the OOP wrapper.

_.mixin({
  capitalize : function(string) {
    return string.charAt(0).toUpperCase() + string.substring(1).toLowerCase();
  }
});
_("fabio").capitalize();
=> "Fabio"

uniqueId_.uniqueId([prefix])
Generate a globally-unique id for client-side models or DOM elements that need one. If prefix is passed, the id will be appended to it. Without prefix, returns an integer.

_.uniqueId('contact_');
=> 'contact_104'

escape_.escape(string)
Escapes a string for insertion into HTML, replacing &, <, >, ", ', and / characters.

_.escape('Curly, Larry & Moe');
=> "Curly, Larry & Moe"

result_.result(object, property)
If the value of the named property is a function then invoke it; otherwise, return it.

var object = {cheese: 'crumpets', stuff: function(){ return 'nonsense'; }};
_.result(object, 'cheese');
=> "crumpets"
_.result(object, 'stuff');
=> "nonsense"

template_.template(templateString, [data], [settings])
Compiles JavaScript templates into functions that can be evaluated for rendering. Useful for rendering complicated bits of HTML from JSON data sources. Template functions can both interpolate variables, using <%= … %>, as well as execute arbitrary JavaScript code, with <% … %>. If you wish to interpolate a value, and have it be HTML-escaped, use <%- … %> When you evaluate a template function, pass in a data object that has properties corresponding to the template's free variables. If you're writing a one-off, you can pass the data object as the second parameter to template in order to render immediately instead of returning a template function. The settings argument should be a hash containing any _.templateSettings that should be overridden.

var compiled = _.template("hello: <%= name %>");
compiled({name : 'moe'});
=> "hello: moe"

var list = "<% _.each(people, function(name) { %> <li><%= name %></li> <% }); %>";
_.template(list, {people : ['moe', 'curly', 'larry']});
=> "<li>moe</li><li>curly</li><li>larry</li>"

var template = _.template("<b><%- value %></b>");
template({value : '<script>'});
=> "<b>&lt;script&gt;</b>"

You can also use print from within JavaScript code. This is sometimes more convenient than using <%= ... %>.

var compiled = _.template("<% print('Hello ' + epithet); %>");
compiled({epithet: "stooge"});
=> "Hello stooge."

If ERB-style delimiters aren't your cup of tea, you can change Underscore's template settings to use different symbols to set off interpolated code. Define an interpolate regex to match expressions that should be interpolated verbatim, an escape regex to match expressions that should be inserted after being HTML escaped, and an evaluate regex to match expressions that should be evaluated without insertion into the resulting string. You may define or omit any combination of the three. For example, to perform Mustache.js style templating:

_.templateSettings = {
  interpolate : /\{\{(.+?)\}\}/g
};

var template = _.template("Hello {{ name }}!");
template({name : "Mustache"});
=> "Hello Mustache!"

By default, template places the values from your data in the local scope via the with statement. However, you can specify a single variable name with the variable setting. This can significantly improve the speed at which a template is able to render.

_.template("Using 'with': <%= data.answer %>", {answer: 'no'}, {variable: 'data'});
=> "Using 'with': no"

Precompiling your templates can be a big help when debugging errors you can't reproduce. This is because precompiled templates can provide line numbers and a stack trace, something that is not possible when compiling templates on the client. The source property is available on the compiled template function for easy precompilation.

<script>
  JST.project = <%= _.template(jstText).source %>;
</script>

Chaining

You can use Underscore in either an object-oriented or a functional style, depending on your preference. The following two lines of code are identical ways to double a list of numbers.

_.map([1, 2, 3], function(n){ return n * 2; });
_([1, 2, 3]).map(function(n){ return n * 2; });

Calling chain will cause all future method calls to return wrapped objects. When you've finished the computation, use value to retrieve the final value. Here's an example of chaining together a map/flatten/reduce, in order to get the word count of every word in a song.

var lyrics = [
  {line : 1, words : "I'm a lumberjack and I'm okay"},
  {line : 2, words : "I sleep all night and I work all day"},
  {line : 3, words : "He's a lumberjack and he's okay"},
  {line : 4, words : "He sleeps all night and he works all day"}
];

_.chain(lyrics)
  .map(function(line) { return line.words.split(' '); })
  .flatten()
  .reduce(function(counts, word) {
    counts[word] = (counts[word] || 0) + 1;
    return counts;
}, {}).value();

=> {lumberjack : 2, all : 4, night : 2 ... }

In addition, the Array prototype's methods are proxied through the chained Underscore object, so you can slip a reverse or a push into your chain, and continue to modify the array.

chain_.chain(obj)
Returns a wrapped object. Calling methods on this object will continue to return wrapped objects until value is used.

var stooges = [{name : 'curly', age : 25}, {name : 'moe', age : 21}, {name : 'larry', age : 23}];
var youngest = _.chain(stooges)
  .sortBy(function(stooge){ return stooge.age; })
  .map(function(stooge){ return stooge.name + ' is ' + stooge.age; })
  .first()
  .value();
=> "moe is 21"

value_(obj).value()
Extracts the value of a wrapped object.

_([1, 2, 3]).value();
=> [1, 2, 3]

Links & Suggested Reading

The Underscore documentation is also available in Simplified Chinese.

Underscore.lua, a Lua port of the functions that are applicable in both languages. Includes OOP-wrapping and chaining. (source)

Underscore.m, an Objective-C port of many of the Underscore.js functions, using a syntax that encourages chaining. (source)

_.m, an alternative Objective-C port that tries to stick a little closer to the original Underscore.js API. (source)

Underscore.php, a PHP port of the functions that are applicable in both languages. Includes OOP-wrapping and chaining. (source)

Underscore-perl, a Perl port of many of the Underscore.js functions, aimed at on Perl hashes and arrays. (source)

Underscore.cfc, a Coldfusion port of many of the Underscore.js functions. (source)

Underscore.string, an Underscore extension that adds functions for string-manipulation: trim, startsWith, contains, capitalize, reverse, sprintf, and more.

Ruby's Enumerable module.

Prototype.js, which provides JavaScript with collection functions in the manner closest to Ruby's Enumerable.

Oliver Steele's Functional JavaScript, which includes comprehensive higher-order function support as well as string lambdas.

Michael Aufreiter's Data.js, a data manipulation + persistence library for JavaScript.

Python's itertools.

Change Log

1.4.2 — Oct. 1, 2012 — Diff

• For backwards compatibility, returned to pre-1.4.0 behavior when passing null to iteration functions. They now become no-ops again.

1.4.1 — Oct. 1, 2012 — Diff

• Fixed a 1.4.0 regression in the lastIndexOf function.

1.4.0 — Sept. 27, 2012 — Diff

• Added a pairs function, for turning a JavaScript object into [key, value] pairs ... as well as an object function, for converting an array of [key, value] pairs into an object.
• Added a countBy function, for counting the number of objects in a list that match a certain criteria.
• Added an invert function, for performing a simple inversion of the keys and values in an object.
• Added a where function, for easy cases of filtering a list for objects with specific values.
• Added an omit function, for filtering an object to remove certain keys.
• Added a random function, to return a random number in a given range.
• _.debounce'd functions now return their last updated value, just like _.throttle'd functions do.
• The sortBy function now runs a stable sort algorithm.
• Added the optional fromIndex option to indexOf and lastIndexOf.
• "Sparse" arrays are no longer supported in Underscore iteration functions. Use a for loop instead (or better yet, an object).
• The min and max functions may now be called on very large arrays.
• Interpolation in templates now represents null and undefined as the empty string.
• Underscore iteration functions no longer accept null values as a no-op argument. You'll get an early error instead.
• A number of edge-cases fixes and tweaks, which you can spot in the diff. Depending on how you're using Underscore, 1.4.0 may be more backwards-incompatible than usual — please test when you upgrade.

1.3.3 — April 10, 2012

• Many improvements to _.template, which now provides the source of the template function as a property, for potentially even more efficient pre-compilation on the server-side. You may now also set the variable option when creating a template, which will cause your passed-in data to be made available under the variable you named, instead of using a with statement — significantly improving the speed of rendering the template.
• Added the pick function, which allows you to filter an object literal with a whitelist of allowed property names.
• Added the result function, for convenience when working with APIs that allow either functions or raw properties.
• Added the isFinite function, because sometimes knowing that a value is a number just ain't quite enough.
• The sortBy function may now also be passed the string name of a property to use as the sort order on each object.
• Fixed uniq to work with sparse arrays.
• The difference function now performs a shallow flatten instead of a deep one when computing array differences.
• The debounce function now takes an immediate parameter, which will cause the callback to fire on the leading instead of the trailing edge.

1.3.1 — Jan. 23, 2012

• Added an _.has function, as a safer way to use hasOwnProperty.
• Added _.collect as an alias for _.map. Smalltalkers, rejoice.
• Reverted an old change so that _.extend will correctly copy over keys with undefined values again.
• Bugfix to stop escaping slashes within interpolations in _.template.

1.3.0 — Jan. 11, 2012

• Removed AMD (RequireJS) support from Underscore. If you'd like to use Underscore with RequireJS, you can load it as a normal script, wrap or patch your copy, or download a forked version.

1.2.4 — Jan. 4, 2012

• You now can (and probably should, as it's simpler) write _.chain(list) instead of _(list).chain().
• Fix for escaped characters in Underscore templates, and for supporting customizations of _.templateSettings that only define one or two of the required regexes.
• Fix for passing an array as the first argument to an _.wrap'd function.
• Improved compatibility with ClojureScript, which adds a call function to String.prototype.

1.2.3 — Dec. 7, 2011

• Dynamic scope is now preserved for compiled _.template functions, so you can use the value of this if you like.
• Sparse array support of _.indexOf, _.lastIndexOf.
• Both _.reduce and _.reduceRight can now be passed an explicitly undefined value. (There's no reason why you'd want to do this.)

1.2.2 — Nov. 14, 2011

• Continued tweaks to _.isEqual semantics. Now JS primitives are considered equivalent to their wrapped versions, and arrays are compared by their numeric properties only (#351).
• _.escape no longer tries to be smart about not double-escaping already-escaped HTML entities. Now it just escapes regardless (#350).
• In _.template, you may now leave semicolons out of evaluated statements if you wish: <% }) %> (#369).
• _.after(callback, 0) will now trigger the callback immediately, making "after" easier to use with asynchronous APIs (#366).

1.2.1 — Oct. 24, 2011

• Several important bug fixes for _.isEqual, which should now do better on mutated Arrays, and on non-Array objects with length properties. (#329)
• jrburke contributed Underscore exporting for AMD module loaders, and tonylukasavage for Appcelerator Titanium. (#335, #338)
• You can now _.groupBy(list, 'property') as a shortcut for grouping values by a particular common property.
• _.throttle'd functions now fire immediately upon invocation, and are rate-limited thereafter (#170, #266).
• Most of the _.is[Type] checks no longer ducktype.
• The _.bind function now also works on constructors, a-la ES5 ... but you would never want to use _.bind on a constructor function.
• _.clone no longer wraps non-object types in Objects.
• _.find and _.filter are now the preferred names for _.detect and _.select.

1.2.0 — Oct. 5, 2011

• The _.isEqual function now supports true deep equality comparisons, with checks for cyclic structures, thanks to Kit Cambridge.
• Underscore templates now support HTML escaping interpolations, using <%- ... %> syntax.
• Ryan Tenney contributed _.shuffle, which uses a modified Fisher-Yates to give you a shuffled copy of an array.
• _.uniq can now be passed an optional iterator, to determine by what criteria an object should be considered unique.
• _.last now takes an optional argument which will return the last N elements of the list.
• A new _.initial function was added, as a mirror of _.rest, which returns all the initial values of a list (except the last N).

1.1.7 — July 13, 2011
Added _.groupBy, which aggregates a collection into groups of like items. Added _.union and _.difference, to complement the (re-named) _.intersection. Various improvements for support of sparse arrays. _.toArray now returns a clone, if directly passed an array. _.functions now also returns the names of functions that are present in the prototype chain.

1.1.6 — April 18, 2011
Added _.after, which will return a function that only runs after first being called a specified number of times. _.invoke can now take a direct function reference. _.every now requires an iterator function to be passed, which mirrors the ECMA5 API. _.extend no longer copies keys when the value is undefined. _.bind now errors when trying to bind an undefined value.

1.1.5 — Mar 20, 2011
Added an _.defaults function, for use merging together JS objects representing default options. Added an _.once function, for manufacturing functions that should only ever execute a single time. _.bind now delegates to the native ECMAScript 5 version, where available. _.keys now throws an error when used on non-Object values, as in ECMAScript 5. Fixed a bug with _.keys when used over sparse arrays.

1.1.4 — Jan 9, 2011
Improved compliance with ES5's Array methods when passing null as a value. _.wrap now correctly sets this for the wrapped function. _.indexOf now takes an optional flag for finding the insertion index in an array that is guaranteed to already be sorted. Avoiding the use of .callee, to allow _.isArray to work properly in ES5's strict mode.

1.1.3 — Dec 1, 2010
In CommonJS, Underscore may now be required with just:
var _ = require("underscore"). Added _.throttle and _.debounce functions. Removed _.breakLoop, in favor of an ECMA5-style un-break-able each implementation — this removes the try/catch, and you'll now have better stack traces for exceptions that are thrown within an Underscore iterator. Improved the isType family of functions for better interoperability with Internet Explorer host objects. _.template now correctly escapes backslashes in templates. Improved _.reduce compatibility with the ECMA5 version: if you don't pass an initial value, the first item in the collection is used. _.each no longer returns the iterated collection, for improved consistency with ES5's forEach.

1.1.2
Fixed _.contains, which was mistakenly pointing at _.intersect instead of _.include, like it should have been. Added _.unique as an alias for _.uniq.

1.1.1
Improved the speed of _.template, and its handling of multiline interpolations. Ryan Tenney contributed optimizations to many Underscore functions. An annotated version of the source code is now available.

1.1.0
The method signature of _.reduce has been changed to match the ECMAScript 5 signature, instead of the Ruby/Prototype.js version. This is a backwards-incompatible change. _.template may now be called with no arguments, and preserves whitespace. _.contains is a new alias for _.include.

1.0.4
Andri Möll contributed the _.memoize function, which can be used to speed up expensive repeated computations by caching the results.

1.0.3
Patch that makes _.isEqual return false if any property of the compared object has a NaN value. Technically the correct thing to do, but of questionable semantics. Watch out for NaN comparisons.

1.0.2
Fixes _.isArguments in recent versions of Opera, which have arguments objects as real Arrays.

1.0.1
Bugfix for _.isEqual, when comparing two objects with the same number of undefined keys, but with different names.

1.0.0
Things have been stable for many months now, so Underscore is now considered to be out of beta, at 1.0. Improvements since 0.6 include _.isBoolean, and the ability to have _.extend take multiple source objects.

0.6.0
Major release. Incorporates a number of Mile Frawley's refactors for safer duck-typing on collection functions, and cleaner internals. A new _.mixin method that allows you to extend Underscore with utility functions of your own. Added _.times, which works the same as in Ruby or Prototype.js. Native support for ECMAScript 5's Array.isArray, and Object.keys.

0.5.8
Fixed Underscore's collection functions to work on NodeLists and HTMLCollections once more, thanks to Justin Tulloss.

0.5.7
A safer implementation of _.isArguments, and a faster _.isNumber,
thanks to Jed Schmidt.

0.5.6
Customizable delimiters for _.template, contributed by Noah Sloan.

0.5.5
Fix for a bug in MobileSafari's OOP-wrapper, with the arguments object.

0.5.4
Fix for multiple single quotes within a template string for _.template. See: Rick Strahl's blog post.

0.5.2
New implementations of isArray, isDate, isFunction, isNumber, isRegExp, and isString, thanks to a suggestion from Robert Kieffer. Instead of doing Object#toString comparisons, they now check for expected properties, which is less safe, but more than an order of magnitude faster. Most other Underscore functions saw minor speed improvements as a result. Evgeniy Dolzhenko contributed _.tap, similar to Ruby 1.9's, which is handy for injecting side effects (like logging) into chained calls.

0.5.1
Added an _.isArguments function. Lots of little safety checks and optimizations contributed by Noah Sloan and Andri Möll.

0.5.0
[API Changes] _.bindAll now takes the context object as its first parameter. If no method names are passed, all of the context object's methods are bound to it, enabling chaining and easier binding. _.functions now takes a single argument and returns the names of its Function properties. Calling _.functions(_) will get you the previous behavior. Added _.isRegExp so that isEqual can now test for RegExp equality. All of the "is" functions have been shrunk down into a single definition. Karl Guertin contributed patches.

0.4.7
Added isDate, isNaN, and isNull, for completeness. Optimizations for isEqual when checking equality between Arrays or Dates. _.keys is now 25%–2X faster (depending on your browser) which speeds up the functions that rely on it, such as _.each.

0.4.6
Added the range function, a port of the Python function of the same name, for generating flexibly-numbered lists of integers. Original patch contributed by Kirill Ishanov.

0.4.5
Added rest for Arrays and arguments objects, and aliased first as head, and rest as tail, thanks to Luke Sutton's patches. Added tests ensuring that all Underscore Array functions also work on arguments objects.

0.4.4
Added isString, and isNumber, for consistency. Fixed _.isEqual(NaN, NaN) to return true (which is debatable).

0.4.3
Started using the native StopIteration object in browsers that support it. Fixed Underscore setup for CommonJS environments.

0.4.2
Renamed the unwrapping function to value, for clarity.

0.4.1
Chained Underscore objects now support the Array prototype methods, so that you can perform the full range of operations on a wrapped array without having to break your chain. Added a breakLoop method to break in the middle of any Underscore iteration. Added an isEmpty function that works on arrays and objects.

0.4.0
All Underscore functions can now be called in an object-oriented style, like so: _([1, 2, 3]).map(...);. Original patch provided by Marc-André Cournoyer. Wrapped objects can be chained through multiple method invocations. A functions method was added, providing a sorted list of all the functions in Underscore.

0.3.3
Added the JavaScript 1.8 function reduceRight. Aliased it as foldr, and aliased reduce as foldl.

0.3.2
Now runs on stock Rhino interpreters with: load("underscore.js"). Added identity as a utility function.

0.3.1
All iterators are now passed in the original collection as their third argument, the same as JavaScript 1.6's forEach. Iterating over objects is now called with (value, key, collection), for details see _.each.

0.3.0
Added Dmitry Baranovskiy's comprehensive optimizations, merged in Kris Kowal's patches to make Underscore CommonJS and Narwhal compliant.

0.2.0
Added compose and lastIndexOf, renamed inject to reduce, added aliases for inject, filter, every, some, and forEach.

0.1.1
Added noConflict, so that the "Underscore" object can be assigned to other variables.

0.1.0
Initial release of Underscore.js.