稽核函数（数组或对象）

each_.each(list, iteratee, [context]) 别名： forEach
遍历list中的所有元素，按顺序用遍历输出每个元素。如果传递了context参数，则把iteratee绑定到context对象上。每次调用iteratee都会传递三个参数：(element, index, list)。如果list是个JavaScript对象，iteratee的参数是 (value, key, list))。返回list以方便链式调用。（注：如果存在原生的forEach方法，Underscore就使用它代替。）

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

注意：集合函数能在数组，对象，和类数组对象，比如arguments, NodeList和类似的数据类型上正常工作。 但是它通过鸭子类型工作，所以要避免传递一个不固定length属性的对象（注：对象或数组的长度（length）属性要固定的）。每个循环不能被破坏 - 打破， 使用_.find代替，这也是很好的注意。

map_.map(list, iteratee, [context]) 别名： collect
通过变换函数（iteratee迭代器）把list中的每个值映射到一个新的数组中（注：产生一个新的数组）。如果存在原生的map方法，就用原生map方法来代替。如果list是个JavaScript对象，iteratee的参数是(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, iteratee, [memo], [context]) Aliases: inject, foldl
别名为 inject 和 foldl, reduce方法把list中元素归结为一个单独的数值。Memo是reduce函数的初始值，reduce的每一步都需要由iteratee返回。这个迭代传递4个参数：memo, value 和 迭代的index（或者 key）和最后一个引用的整个 list。

如果没有memo传递给reduce的初始调用，iteratee不会被列表中的第一个元素调用。第一个元素将取代 传递给列表中下一个元素调用iteratee的memo参数，

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

reduceRight_.reduceRight(list, iteratee, memo, [context]) 别名: foldr
reducRight是从右侧开始组合的元素的reduce函数，如果存在JavaScript 1.8版本的reduceRight，则用其代替。Foldr在javascript中不像其它有懒计算的语言那么有用（注：lazy evaluation：一种求值策略，只有当表达式的值真正需要时才对表达式进行计算）。

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, predicate, [context]) 别名： detect
在list中逐项查找，返回第一个通过predicate迭代函数真值检测的元素值，如果没有值传递给测试迭代器将返回undefined。 如果找到匹配的元素，函数将立即返回，不会遍历整个list。

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

filter_.filter(list, predicate, [context]) 别名： select
遍历list中的每个值，返回包含所有通过predicate真值检测的元素值。（注：如果存在原生filter方法，则用原生的filter方法。）

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

where_.where(list, properties)
遍历list中的每一个值，返回一个数组，这个数组包含包含properties所列出的属性的所有的键 - 值对。

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

findWhere_.findWhere(list, properties)
遍历list中的每一个值，返回匹配properties所列出的属性的所有的键 - 值对的第一个值。

如果没有找到匹配的属性，或者list是空的，那么将返回undefined。

_.findWhere(publicServicePulitzers, {newsroom: "The New York Times"});
=> {year: 1918, newsroom: "The New York Times",
  reason: "For its public service in publishing in full so many official reports,
  documents and speeches by European statesmen relating to the progress and
  conduct of the war."}

reject_.reject(list, predicate, [context])
返回list中没有通过predicate真值检测的元素集合，与filter相反。

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

every_.every(list, [predicate], [context]) 别名： all
如果list中的所有元素都通过predicate的真值检测就返回true。（注：如果存在原生的every方法，就使用原生的every。）

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

some_.some(list, [predicate], [context]) 别名： any
如果list中有任何一个元素通过 predicate 的真值检测就返回true。一旦找到了符合条件的元素, 就直接中断对list的遍历. （注：如果存在原生的some方法，就使用原生的some。）

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

contains_.contains(list, value) 别名： include
如果list包含指定的value则返回true（注：使用===检测）。如果list 是数组，内部使用indexOf判断。

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

invoke_.invoke(list, methodName, *arguments)
在list的每个元素上执行methodName方法。 任何传递给invoke的额外参数，invoke都会在调用methodName方法的时候传递给它。

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

pluck_.pluck(list, propertyName)
pluck也许是map最常使用的用例模型的简化版本，即萃取对象数组中某属性值，返回一个数组。

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

max_.max(list, [iteratee], [context])
返回list中的最大值。如果传递iteratee参数，iteratee将作为list中每个值的排序依据。如果list为空，将返回-Infinity，所以你可能需要事先用isEmpty检查 list 。

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, [iteratee], [context])
返回list中的最小值。如果传递iteratee参数，iteratee将作为list中每个值的排序依据。如果list为空，将返回-Infinity，所以你可能需要事先用isEmpty检查 list 。

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

sortBy_.sortBy(list, iteratee, [context])
返回一个排序后的list拷贝副本。如果传递iteratee参数，iteratee将作为list中每个值的排序依据。迭代器也可以是字符串的属性的名称进行排序的(比如 length)。

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

groupBy_.groupBy(list, iteratee, [context])
把一个集合分组为多个集合，通过 iterator 返回的结果进行分组. 如果 iterator 是一个字符串而不是函数, 那么将使用 iterator 作为各元素的属性名来对比进行分组.

_.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"]}

indexBy_.indexBy(list, iteratee, [context])
给定一个list，和 一个用来返回一个在列表中的每个元素键 的iterator 函数（或属性名）， 返回一个每一项索引的对象。和groupBy非常像，但是当你知道你的键是唯一的时候可以使用indexBy 。

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

countBy_.countBy(list, iteratee, [context])
排序一个列表组成一个组，并且返回各组中的对象的数量的计数。类似groupBy，但是不是返回列表的值，而是返回在该组中值的数目。

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

shuffle_.shuffle(list)
返回一个随机乱序的 list 副本, 使用 Fisher-Yates shuffle 来进行随机乱序.

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

sample_.sample(list, [n])
从 list中产生一个随机样本。传递一个数字表示从list中返回n个随机元素。否则将返回一个单一的随机项。

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

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

toArray_.toArray(list)
把list(任何可以迭代的对象)转换成一个数组，在转换 arguments 对象时非常有用。

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

size_.size(list)
返回list的长度。

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

partition_.partition(array, predicate)
拆分一个数组（array）为两个数组： 第一个数组其元素都满足predicate迭代函数， 而第二个的所有元素均不能满足predicate迭代函数。

_.partition([0, 1, 2, 3, 4, 5], isOdd);
=> [[1, 3, 5], [0, 2, 4]]

数组函数（Array Functions）

注： arguments（参数） 对象将在所有数组函数中工作 。然而, Underscore 函数的设计并不只是针对稀疏（"sparse" ）数组的.

first_.first(array, [n]) 别名： head, take
返回array（数组）的第一个元素。传递 n参数将返回数组中从第一个元素开始的n个元素（注：返回数组中前 n 个元素.）。

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

initial_.initial(array, [n])
返回数组中除了最后一个元素外的其他全部元素。 在arguments对象上特别有用。传递 n参数将从结果中排除从最后一个开始的n个元素（注：排除数组后面的 n 个元素）。

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

last_.last(array, [n])
返回array（数组）的最后一个元素。传递 n参数将返回数组中从最后一个元素开始的n个元素（注：返回数组里的后面的n个元素）。

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

rest_.rest(array, [index]) 别名： tail, drop
返回数组中除了第一个元素外的其他全部元素。传递 index 参数将返回从index开始的剩余所有元素 。（感谢@德德德德撸 指出错误）

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

compact_.compact(array)
返回一个除去所有false值的 array副本。 在javascript中, false, null, 0, "", undefined 和 NaN 都是false值.

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

flatten_.flatten(array, [shallow])
将一个嵌套多层的数组 array（数组） (嵌套可以是任何层数)转换为只有一层的数组。 如果你传递 shallow参数，数组将只减少一维的嵌套。

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

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

without_.without(array, *values)
返回一个删除所有values值后的 array副本。（注：使用===表达式做相等测试。）

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

union_.union(*arrays)
返回传入的 arrays（数组）并集：按顺序返回，返回数组的元素是唯一的，可以传入一个或多个 arrays（数组）。

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

intersection_.intersection(*arrays)
返回传入 arrays（数组）交集。结果中的每个值是存在于传入的每个arrays（数组）里。

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

difference_.difference(array, *others)
类似于without，但返回的值来自array参数数组，并且不存在于other 数组.

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

uniq_.uniq(array, [isSorted], [iteratee]) 别名： unique
返回 array去重后的副本, 使用 === 做相等测试. 如果您确定 array 已经排序, 那么给 isSorted 参数传递 true值, 此函数将运行的更快的算法. 如果要处理对象元素, 传参 iterator 来获取要对比的属性.

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

zip_.zip(*arrays)
将 每个arrays中相应位置的值合并在一起。在合并分开保存的数据时很有用. 如果你用来处理矩阵嵌套数组时, _.zip.apply 可以做类似的效果。

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

_.zip.apply(_, arrayOfRowsOfData);
=> arrayOfColumnsOfData

object_.object(list, [values])
将数组转换为对象。传递任何一个单独[key, value]对的列表，或者一个键的列表和一个值得列表。 如果存在重复键，最后一个值将被返回。

_.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])
返回value在该 array 中的索引值，如果value不存在 array中就返回-1。使用原生的indexOf 函数，除非它失效。如果您正在使用一个大数组，你知道数组已经排序，传递true给isSorted将更快的用二进制搜索..,或者，传递一个数字作为第三个参数，为了在给定的索引的数组中寻找第一个匹配值。

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

lastIndexOf_.lastIndexOf(array, value, [fromIndex])
返回value在该 array 中的从最后开始的索引值，如果value不存在 array中就返回-1。如果支持原生的lastIndexOf，将使用原生的lastIndexOf函数。 传递fromIndex将从你给定的索性值开始搜索。

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

sortedIndex_.sortedIndex(list, value, [iteratee], [context])
使用二分查找确定value在list中的位置序号，value按此序号插入能保持list原有的排序。 如果提供iterator函数，iterator将作为list排序的依据，包括你传递的value 。 iterator也可以是字符串的属性名用来排序(比如length)。

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

var stooges = [{name: 'moe', age: 40}, {name: 'curly', age: 60}];
_.sortedIndex(stooges, {name: 'larry', age: 50}, 'age');
=> 1

range_.range([start], stop, [step])
一个用来创建整数灵活编号的列表的函数，便于each 和 map循环。如果省略start则默认为 0；step 默认为 1.返回一个从start 到stop的整数的列表，用step来增加 （或减少）独占。值得注意的是，如果stop值在start前面（也就是stop值小于start值），那么值域会被认为是零长度，而不是负增长。-如果你要一个负数的值域 ，请使用负数step.

_.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);
=> []

与函数有关的函数（Function (uh, ahem) Functions）

bind_.bind(function, object, *arguments)
绑定函数 function 到对象 object 上, 也就是无论何时调用函数, 函数里的 this 都指向这个 object. 任意可选参数 arguments 可以传递给函数 function , 可以填充函数所需要的参数, 这也被称为 partial application。对于没有结合上下文的partial application绑定，请使用partial。
(注：partial application翻译成“部分应用”或者“偏函数应用”。partial application可以被描述为一个函数，它接受一定数目的参数，绑定值到一个或多个这些参数，并返回一个新的函数，这个返回函数只接受剩余未绑定值的参数。参见：http://en.wikipedia.org/wiki/Partial_application。感谢@一任风月忆秋年的建议)。

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

bindAll_.bindAll(object, *methodNames)
把methodNames参数指定的一些方法绑定到object上，这些方法就会在对象的上下文环境中执行。绑定函数用作事件处理函数时非常便利，否则函数被调用时this一点用也没有。methodNames参数是必须的。

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

partial_.partial(function, *arguments)
局部应用一个函数填充在任意个数的 参数，不改变其动态this值。和bind方法很相近。你可以在你的参数列表中传递_来指定一个参数 ，不应该被预先填充， but left open to supply at call-time. You may pass _ in your list of arguments to specify an argument that should not be pre-filled, but left open to supply at call-time.（翻译不好，求好的翻译）

var add = function(a, b) { return a + b; };
add5 = _.partial(add, 5);
add5(10);
=> 15

memoize_.memoize(function, [hashFunction])
Memoizes方法可以缓存某函数的计算结果。对于耗时较长的计算是很有帮助的。如果传递了 hashFunction 参数，就用 hashFunction 的返回值作为key存储函数的计算结果。 hashFunction 默认使用function的第一个参数作为key。memoized值的缓存 可作为 返回函数的cache属性。

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

delay_.delay(function, wait, *arguments)
类似setTimeout，等待wait毫秒后调用function。如果传递可选的参数arguments，当函数function执行时， arguments 会作为参数传入。

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

defer_.defer(function, *arguments)
延迟调用function直到当前调用栈清空为止，类似使用延时为0的setTimeout方法。对于执行开销大的计算和无阻塞UI线程的HTML渲染时候非常有用。 如果传递arguments参数，当函数function执行时， arguments 会作为参数传入。

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

throttle_.throttle(function, wait, [options])
创建并返回一个像节流阀一样的函数，当重复调用函数的时候，最多每隔 wait毫秒调用一次该函数。对于想控制一些触发频率较高的事件有帮助。（注：详见：javascript函数的throttle和debounce）

默认情况下，throttle将在你调用的第一时间尽快执行这个function，并且，如果你在wait周期内调用任意次数的函数，都将尽快的被覆盖。如果你想禁用第一次首先执行的话，传递{leading: false}，还有如果你想禁用最后一次执行的话，传递{trailing: false}。

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

debounce_.debounce(function, wait, [immediate])
返回 function 函数的防反跳版本, 将延迟函数的执行(真正的执行)在函数最后一次调用时刻的 wait 毫秒之后. 对于必须在一些输入（多是一些用户操作）停止到达之后执行的行为有帮助。 例如: 渲染一个Markdown格式的评论预览, 当窗口停止改变大小之后重新计算布局, 等等.

传参 immediate 为 true， debounce会在 wait 时间间隔的开始调用这个函数 。（注：并且在 waite 的时间之内，不会再次调用。）在类似不小心点了提交按钮两下而提交了两次的情况下很有用。 （感谢 @ProgramKid 的翻译建议）

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

once_.once(function)
创建一个只能调用一次的函数。重复调用改进的方法也没有效果，只会返回第一次执行时的结果。 作为初始化函数使用时非常有用, 不用再设一个boolean值来检查是否已经初始化完成.

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

after_.after(count, function)
创建一个函数, 只有在运行了 count 次之后才有效果. 在处理同组异步请求返回结果时, 如果你要确保同组里所有异步请求完成之后才 执行这个函数, 这将非常有用。

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

before_.before(count, function)
创建一个函数,调用不超过count 次。 当count已经达到时，最后一个函数调用的结果 是被记住并返回 。

var monthlyMeeting = _.before(3, askForRaise);
monthlyMeeting();
monthlyMeeting();
monthlyMeeting();
// the result of any subsequent calls is the same as the second call

wrap_.wrap(function, wrapper)
将第一个函数 function 封装到函数 wrapper 里面, 并把函数 function 作为第一个参数传给 wrapper. 这样可以让 wrapper 在 function 运行之前和之后 执行代码, 调整参数然后附有条件地执行.

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

negate_.negate(predicate)
返回一个新的predicate函数的否定版本。

var isFalsy = _.negate(Boolean);
_.find([-2, -1, 0, 1, 2], isFalsy);
=> 0

compose_.compose(*functions)
返回函数集 functions 组合后的复合函数, 也就是一个函数执行完之后把返回的结果再作为参数赋给下一个函数来执行. 以此类推. 在数学里, 把函数 f(), g(), 和 h() 组合起来可以得到复合函数 f(g(h()))。

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

对象函数（Object Functions）

keys_.keys(object)
获取object对象所有的属性名称。

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

values_.values(object)
返回object对象所有的属性值。

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

pairs_.pairs(object)
把一个对象转变为一个[key, value]形式的数组。

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

invert_.invert(object)
返回一个object副本，使其键（keys）和值（values）对换。对于这个操作，必须确保object里所有的值都是唯一的且可以序列号成字符串.

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

functions_.functions(object) 别名： methods
返回一个对象里所有的方法名, 而且是已经排序的 — 也就是说, 对象里每个方法(属性值是一个函数)的名称.

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

extend_.extend(destination, *sources)
复制source对象中的所有属性覆盖到destination对象上，并且返回 destination 对象. 复制是按顺序的, 所以后面的对象属性会把前面的对象属性覆盖掉(如果有重复).

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

pick_.pick(object, *keys)
返回一个object副本，只过滤出keys(有效的键组成的数组)参数指定的属性值。或者接受一个判断函数，指定挑选哪个key。

_.pick({name: 'moe', age: 50, userid: 'moe1'}, 'name', 'age');
=> {name: 'moe', age: 50}
_.pick({name: 'moe', age: 50, userid: 'moe1'}, function(value, key, object) {
  return _.isNumber(value);
});
=> {age: 50}

omit_.omit(object, *keys)
返回一个object副本，只过滤出除去keys(有效的键组成的数组)参数指定的属性值。 或者接受一个判断函数，指定忽略哪个key。

_.omit({name: 'moe', age: 50, userid: 'moe1'}, 'userid');
=> {name: 'moe', age: 50}
_.omit({name: 'moe', age: 50, userid: 'moe1'}, function(value, key, object) {
  return _.isNumber(value);
});
=> {name: 'moe', userid: 'moe1'}

defaults_.defaults(object, *defaults)
用defaults对象填充object 中的undefined属性。 并且返回这个object。一旦这个属性被填充，再使用defaults方法将不会有任何效果。（感谢@一任风月忆秋年的拍砖）

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

clone_.clone(object)
创建 一个浅复制（浅拷贝）的克隆object。任何嵌套的对象或数组都通过引用拷贝，不会复制。

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

tap_.tap(object, interceptor)
用 object作为参数来调用函数interceptor，然后返回object。这种方法的主要意图是作为函数链式调用 的一环, 为了对此对象执行操作并返回对象本身。

_.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)
对象是否包含给定的键吗？等同于object.hasOwnProperty(key)，但是使用hasOwnProperty 函数的一个安全引用，以防意外覆盖。

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

property_.property(key)
返回一个函数，这个函数返回任何传入的对象的key 属性。

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

matches_.matches(attrs)
返回一个断言函数，这个函数会给你一个断言 可以用来辨别 给定的对象是否匹配attrs指定键/值属性。

var ready = _.matches({selected: true, visible: true});
var readyToGoList = _.filter(list, ready);

isEqual_.isEqual(object, other)
执行两个对象之间的优化深度比较，确定他们是否应被视为相等。

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)
如果object 不包含任何值(没有可枚举的属性)，返回true。 对于字符串和类数组（array-like）对象，如果length属性为0，那么_.isEmpty检查返回true。

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

isElement_.isElement(object)
如果object是一个DOM元素，返回true。

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

isArray_.isArray(object)
如果object是一个数组，返回true。

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

isObject_.isObject(value)
如果object是一个对象，返回true。需要注意的是JavaScript数组和函数是对象，字符串和数字不是。

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

isArguments_.isArguments(object)
如果object是一个参数对象，返回true。

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

isFunction_.isFunction(object)
如果object是一个函数（Function），返回true。

_.isFunction(alert);
=> true

isString_.isString(object)
如果object是一个字符串，返回true。

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

isNumber_.isNumber(object)
如果object是一个数值，返回true (包括 NaN)。

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

isFinite_.isFinite(object)
如果object是一个有限的数字，返回true。

_.isFinite(-101);
=> true

_.isFinite(-Infinity);
=> false

isBoolean_.isBoolean(object)
如果object是一个布尔值，返回true。 Returns true if object is either true or false.

_.isBoolean(null);
=> false

isDate_.isDate(object)
如果object是一个Date类型（日期时间），返回true。

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

isRegExp_.isRegExp(object)
如果object是一个正则表达式，返回true。

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

isNaN_.isNaN(object)
如果object是 NaN，返回true。
注意： 这和原生的isNaN 函数不一样，如果变量是undefined，原生的isNaN 函数也会返回 true 。

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

isNull_.isNull(object)
如果object的值是 null，返回true。

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

isUndefined_.isUndefined(value)
如果value是undefined，返回true。

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

实用功能(Utility Functions)

noConflict_.noConflict()
放弃Underscore 的控制变量"_"。返回Underscore 对象的引用。

var underscore = _.noConflict();

identity_.identity(value)
返回与传入参数相等的值. 相当于数学里的: f(x) = x
这个函数看似无用, 但是在Underscore里被用作默认的迭代器iterator.

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

constant_.constant(value)
创建一个函数，这个函数 返回相同的值 用来作为_.constant的参数。

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

noop_.noop()
返回undefined，不论传递给它的是什么参数。 可以用作默认可选的回调参数。

obj.initialize = _.noop;

times_.times(n, iteratee, [context])
调用给定的迭代函数n次,每一次调用iteratee传递index参数。生成一个返回值的数组。
注意: 本例使用 链式语法。

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

random_.random(min, max)
返回一个min 和 max之间的随机整数。如果你只传递一个参数，那么将返回0和这个参数之间的整数。

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

mixin_.mixin(object)
允许用您自己的实用程序函数扩展Underscore。传递一个 {name: function}定义的哈希添加到Underscore对象，以及面向对象封装。

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

iteratee_.iteratee(value, [context], [argCount])
一个重要的内部函数用来生成可应用到集合中每个元素的回调， 返回想要的结果 - 无论是等式，任意回调，属性匹配，或属性访问。
通过_.iteratee转换判断的Underscore 方法的完整列表是 map, find, filter, reject, every, some, max, min, sortBy, groupBy, indexBy, countBy, sortedIndex, partition, 和 unique.

var stooges = [{name: 'curly', age: 25}, {name: 'moe', age: 21}, {name: 'larry', age: 23}];
_.map(stooges, _.iteratee('age'));
=> [25, 21, 23];

uniqueId_.uniqueId([prefix])
为需要的客户端模型或DOM元素生成一个全局唯一的id。如果prefix参数存在， id 将附加给它。

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

escape_.escape(string)
转义HTML字符串，替换&, <, >, ", ', 和 /字符。

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

unescape_.unescape(string)
和escape相反。转义HTML字符串，替换&, &lt;, &gt;, &quot;, &#96;, 和 &#x2F;字符。

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

result_.result(object, property)
如果对象 object 中的属性 property 是函数, 则调用它, 否则, 返回它。

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

now_.now()
一个优化的方式来获得一个当前时间的整数时间戳。 可用于实现定时/动画功能。

_.now();
=> 1392066795351

template_.template(templateString, [settings])
将 JavaScript 模板编译为可以用于页面呈现的函数, 对于通过JSON数据源生成复杂的HTML并呈现出来的操作非常有用。 模板函数可以使用 <%= … %>插入变量, 也可以用<% … %>执行任意的 JavaScript 代码。 如果您希望插入一个值, 并让其进行HTML转义,请使用<%- … %>。 当你要给模板函数赋值的时候，可以传递一个含有与模板对应属性的data对象 。 如果您要写一个一次性的, 您可以传对象 data 作为第二个参数给模板 template 来直接呈现, 这样页面会立即呈现而不是返回一个模板函数. 参数 settings 是一个哈希表包含任何可以覆盖的设置 _.templateSettings.

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

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

您也可以在JavaScript代码中使用 print. 有时候这会比使用 <%= ... %> 更方便.

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

如果ERB式的分隔符您不喜欢, 您可以改变Underscore的模板设置, 使用别的符号来嵌入代码. 定义一个 interpolate 正则表达式来逐字匹配 嵌入代码的语句, 如果想插入转义后的HTML代码 则需要定义一个 escape 正则表达式来匹配, 还有一个 evaluate 正则表达式来匹配 您想要直接一次性执行程序而不需要任何返回值的语句. 您可以定义或省略这三个的任意一个. 例如, 要执行 Mustache.js 类型的模板:

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

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

默认的, template 通过 with 语句 来取得 data 所有的值. 当然, 您也可以在 variable 设置里指定一个变量名. 这样能显著提升模板的渲染速度.

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

预编译模板对调试不可重现的错误很有帮助. 这是因为预编译的模板可以提供错误的代码行号和堆栈跟踪, 有些模板在客户端(浏览器)上是不能通过编译的 在编译好的模板函数上, 有 source 属性可以提供简单的预编译功能.

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

链式语法(Chaining)

您可以在面向对象或者函数的风格下使用Underscore, 这取决于您的个人偏好. 以下两行代码都可以 把一个数组里的所有数字乘以2.

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

对一个对象使用 chain 方法, 会把这个对象封装并 让以后每次方法的调用结束后都返回这个封装的对象, 当您完成了计算, 可以使用 value 函数来取得最终的值. 以下是一个同时使用了 map/flatten/reduce 的链式语法例子, 目的是计算一首歌的歌词里每一个单词出现的次数.

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 ... }

此外, 数组原型方法 也通过代理加入到了链式封装的Underscore对象, 所以您可以 在链式语法中直接使用 reverse 或 push 方法, 然后再接着其他的语句.

chain_.chain(obj)
返回一个封装的对象. 在封装的对象上调用方法会返回封装的对象本身, 直道 value 方法调用为止.

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()
获取封装对象的最终值.

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

更多链接 & 推荐阅读（Links & Suggested Reading）

Underscore文档也有 简体中文 版

Underscore.lua, 一个Lua版本的Underscore, 函数都通用. 包含面向对象封装和链式语法. (源码)

Underscore.m, 一个 Objective-C 版本的 Underscore.js, 实现了大部分函数, 它的语法鼓励使用链式语法. (源码)

_.m, 另一个 Objective-C 版本, 这个版本与原始的 Underscore.js API 比较相近. (源码)

Underscore.php, 一个PHP版本的Underscore, 函数都通用. 包含面向对象封装和链式语法. (源码)

Underscore-perl, 一个Perl版本的Underscore, 实现了大部分功能, 主要针对于Perl的哈希表和数组. (源码)

Underscore.cfc, 一个 Coldfusion 版本的 Underscore.js, 实现了大部分函数. (源码)

Underscore.string, 一个Underscore的扩展, 添加了多个字符串操作的函数, 如: trim, startsWith, contains, capitalize, reverse, sprintf, 还有更多.

Ruby的 枚举 模块.

Prototype.js, 提供类似于Ruby枚举方式的JavaScript集合函数.

Oliver Steele的 Functional JavaScript, 包含全面的高阶函数支持以及字符串的匿名函数.

Michael Aufreiter的 Data.js, 一个JavaScript的数据操作和持久化的类库.

Python的 迭代工具.

PyToolz, 一个Python端口 扩展了itertools和functools，包括很多的Underscore API。

Funcy, a practical collection of functional helpers for Python, partially inspired by Underscore.

更新日志

1.7.0 — August 26, 2014 — Diff — Docs

• For consistency and speed across browsers, Underscore now ignores native array methods for forEach, map, reduce, reduceRight, filter, every, some, indexOf, and lastIndexOf. "Sparse" arrays are officially dead in Underscore.
• Added _.iteratee to customize the iterators used by collection functions. Many Underscore methods will take a string argument for easier _.property-style lookups, an object for _.where-style filtering, or a function as a custom callback.
• Added _.before as a counterpart to _.after.
• Added _.negate to invert the truth value of a passed-in predicate.
• Added _.noop as a handy empty placeholder function.
• _.isEmpty now works with arguments objects.
• _.has now guards against nullish objects.
• _.omit can now take an iteratee function.
• _.partition is now called with index and object.
• _.matches creates a shallow clone of your object and only iterates over own properties.
• Aligning better with the forthcoming ECMA6 Object.assign, _.extend only iterates over the object's own properties.
• Falsey guards are no longer needed in _.extend and _.defaults—if the passed in argument isn't a JavaScript object it's just returned.
• Fixed a few edge cases in _.max and _.min to handle arrays containing NaN (like strings or other objects) and Infinity and -Infinity.
• Override base methods like each and some and they'll be used internally by other Underscore functions too.
• The escape functions handle backticks (`), to deal with an IE ≤ 8 bug.
• For consistency, _.union and _.difference now only work with arrays and not variadic args.
• _.memoize exposes the cache of memoized values as a property on the returned function.
• _.pick accepts iteratee and context arguments for a more advanced callback.
• Underscore templates no longer accept an initial data object. _.template always returns a function now.
• Optimizations and code cleanup aplenty.

1.6.0 — February 10, 2014 — Diff — Docs

• Underscore now registers itself for AMD (Require.js), Bower and Component, as well as being a CommonJS module and a regular (Java)Script. An ugliness, but perhaps a necessary one.
• Added _.partition, a way to split a collection into two lists of results — those that pass and those that fail a particular predicate.
• Added _.property, for easy creation of iterators that pull specific properties from objects. Useful in conjunction with other Underscore collection functions.
• Added _.matches, a function that will give you a predicate that can be used to tell if a given object matches a list of specified key/value properties.
• Added _.constant, as a higher-order _.identity.
• Added _.now, an optimized way to get a timestamp — used internally to speed up debounce and throttle.
• The _.partial function may now be used to partially apply any of its arguments, by passing _ wherever you'd like a placeholder variable, to be filled-in later.
• The _.each function now returns a reference to the list for chaining.
• The _.keys function now returns an empty array for non-objects instead of throwing.
• … and more miscellaneous refactoring.

1.5.2 — September 7, 2013 — Diff — Docs

• Added an indexBy function, which fits in alongside its cousins, countBy and groupBy.
• Added a sample function, for sampling random elements from arrays.
• Some optimizations relating to functions that can be implemented in terms of _.keys (which includes, significantly, each on objects). Also for debounce in a tight loop.
• The _.escape function no longer escapes '/'.

1.5.1 — July 8, 2013 — Diff — Docs

• Removed unzip, as it's simply the application of zip to an array of arguments. Use _.zip.apply(_, list) to transpose instead.

1.5.0 — July 6, 2013 — Diff — Docs

• Added a new unzip function, as the inverse of _.zip.
• The throttle function now takes an options argument, allowing you to disable execution of the throttled function on either the leading or trailing edge.
• A source map is now supplied for easier debugging of the minified production build of Underscore.
• The defaults function now only overrides undefined values, not null ones.
• Removed the ability to call _.bindAll with no method name arguments. It's pretty much always wiser to white-list the names of the methods you'd like to bind.
• Removed the ability to call _.after with an invocation count of zero. The minimum number of calls is (naturally) now 1.

1.4.4 — January 30, 2013 — Diff — Docs

• Added _.findWhere, for finding the first element in a list that matches a particular set of keys and values.
• Added _.partial, for partially applying a function without changing its dynamic reference to this.
• Simplified bind by removing some edge cases involving constructor functions. In short: don't _.bind your constructors.
• A minor optimization to invoke.
• Fix bug in the minified version due to the minifier incorrectly optimizing-away isFunction.

1.4.3 — December 4, 2012 — Diff — Docs

• Improved Underscore compatibility with Adobe's JS engine that can be used to script Illustrator, Photoshop, and friends.
• Added a default _.identity iterator to countBy and groupBy.
• The uniq function can now take array, iterator, context as the argument list.
• The times function now returns the mapped array of iterator results.
• Simplified and fixed bugs in throttle.

1.4.2 — October 6, 2012 — Diff — Docs

• 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 — October 1, 2012 — Diff — Docs

• Fixed a 1.4.0 regression in the lastIndexOf function.

1.4.0 — September 27, 2012 — Diff — Docs

• 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 — Diff — Docs

• 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 — January 23, 2012 — Diff — Docs

• 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 — January 11, 2012 — Diff — Docs

• 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 — January 4, 2012 — Diff — Docs

• 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 — December 7, 2011 — Diff — Docs

• 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 — November 14, 2011 — Diff — Docs

• 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 — October 24, 2011 — Diff — Docs

• Several important bug fixes for _.isEqual, which should now do better on mutated Arrays, and on non-Array objects with length properties. (#329)
• James Burke contributed Underscore exporting for AMD module loaders, and Tony Lukasavage 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 — October 5, 2011 — Diff — Docs

• 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 — Diff — Docs
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 — Diff — Docs
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 ES5 API. _.extend no longer copies keys when the value is undefined. _.bind now errors when trying to bind an undefined value.

1.1.5 — March 20, 2011 — Diff — Docs
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 ES5 version, where available. _.keys now throws an error when used on non-Object values, as in ES5. Fixed a bug with _.keys when used over sparse arrays.

1.1.4 — January 9, 2011 — Diff — Docs
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 — December 1, 2010 — Diff — Docs
In CommonJS, Underscore may now be required with just:
var _ = require("underscore"). Added _.throttle and _.debounce functions. Removed _.breakLoop, in favor of an ES5-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 ES5 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 — October 15, 2010 — Diff — Docs
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 — October 5, 2010 — Diff — Docs
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 — August 18, 2010 — Diff — Docs
The method signature of _.reduce has been changed to match the ES5 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 — June 22, 2010 — Diff — Docs
Andri Möll contributed the _.memoize function, which can be used to speed up expensive repeated computations by caching the results.

1.0.3 — June 14, 2010 — Diff — Docs
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 — March 23, 2010 — Diff — Docs
Fixes _.isArguments in recent versions of Opera, which have arguments objects as real Arrays.

1.0.1 — March 19, 2010 — Diff — Docs
Bugfix for _.isEqual, when comparing two objects with the same number of undefined keys, but with different names.

1.0.0 — March 18, 2010 — Diff — Docs
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 — February 24, 2010 — Diff — Docs
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 ES5's Array.isArray, and Object.keys.

0.5.8 — January 28, 2010 — Diff — Docs
Fixed Underscore's collection functions to work on NodeLists and HTMLCollections once more, thanks to Justin Tulloss.

0.5.7 — January 20, 2010 — Diff — Docs
A safer implementation of _.isArguments, and a faster _.isNumber,
thanks to Jed Schmidt.

0.5.6 — January 18, 2010 — Diff — Docs
Customizable delimiters for _.template, contributed by Noah Sloan.

0.5.5 — January 9, 2010 — Diff — Docs
Fix for a bug in MobileSafari's OOP-wrapper, with the arguments object.

0.5.4 — January 5, 2010 — Diff — Docs
Fix for multiple single quotes within a template string for _.template. See: Rick Strahl's blog post.

0.5.2 — January 1, 2010 — Diff — Docs
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 — December 9, 2009 — Diff — Docs
Added an _.isArguments function. Lots of little safety checks and optimizations contributed by Noah Sloan and Andri Möll.

0.5.0 — December 7, 2009 — Diff — Docs
[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 — December 6, 2009 — Diff — Docs
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 — November 30, 2009 — Diff — Docs
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 — November 19, 2009 — Diff — Docs
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 — November 18, 2009 — Diff — Docs
Added isString, and isNumber, for consistency. Fixed _.isEqual(NaN, NaN) to return true (which is debatable).

0.4.3 — November 9, 2009 — Diff — Docs
Started using the native StopIteration object in browsers that support it. Fixed Underscore setup for CommonJS environments.

0.4.2 — November 9, 2009 — Diff — Docs
Renamed the unwrapping function to value, for clarity.

0.4.1 — November 8, 2009 — Diff — Docs
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 — November 7, 2009 — Diff — Docs
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 — October 31, 2009 — Diff — Docs
Added the JavaScript 1.8 function reduceRight. Aliased it as foldr, and aliased reduce as foldl.

0.3.2 — October 29, 2009 — Diff — Docs
Now runs on stock Rhino interpreters with: load("underscore.js"). Added identity as a utility function.

0.3.1 — October 29, 2009 — Diff — Docs
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 — October 29, 2009 — Diff — Docs
Added Dmitry Baranovskiy's comprehensive optimizations, merged in Kris Kowal's patches to make Underscore CommonJS and Narwhal compliant.

0.2.0 — October 28, 2009 — Diff — Docs
Added compose and lastIndexOf, renamed inject to reduce, added aliases for inject, filter, every, some, and forEach.

0.1.1 — October 28, 2009 — Diff — Docs
Added noConflict, so that the "Underscore" object can be assigned to other variables.

0.1.0 — October 28, 2009 — Docs
Initial release of Underscore.js.

此“Underscore.js 中文文档”参考了WEB前端开发网的精彩译文。