Most efficient method to groupby on an array of objects

Using ES6 Map object:

/**
 * @description
 * Takes an Array<V>, and a grouping function,
 * and returns a Map of the array grouped by the grouping function.
 *
 * @param list An array of type V.
 * @param keyGetter A Function that takes the the Array type V as an input, and returns a value of type K.
 *                  K is generally intended to be a property key of V.
 *
 * @returns Map of the array grouped by the grouping function.
 */
//export function groupBy<K, V>(list: Array<V>, keyGetter: (input: V) => K): Map<K, Array<V>> {
//    const map = new Map<K, Array<V>>();
function groupBy(list, keyGetter) {
    const map = new Map();
    list.forEach((item) => {
         const key = keyGetter(item);
         const collection = map.get(key);
         if (!collection) {
             map.set(key, [item]);
         } else {
             collection.push(item);
         }
    });
    return map;
}


// example usage

const pets = [
    {type:"Dog", name:"Spot"},
    {type:"Cat", name:"Tiger"},
    {type:"Dog", name:"Rover"}, 
    {type:"Cat", name:"Leo"}
];
    
const grouped = groupBy(pets, pet => pet.type);
    
console.log(grouped.get("Dog")); // -> [{type:"Dog", name:"Spot"}, {type:"Dog", name:"Rover"}]
console.log(grouped.get("Cat")); // -> [{type:"Cat", name:"Tiger"}, {type:"Cat", name:"Leo"}]

const odd = Symbol();
const even = Symbol();
const numbers = [1,2,3,4,5,6,7];

const oddEven = groupBy(numbers, x => (x % 2 === 1 ? odd : even));
    
console.log(oddEven.get(odd)); // -> [1,3,5,7]
console.log(oddEven.get(even)); // -> [2,4,6]

About Map: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Map

with ES6:

const groupBy = (items, key) => items.reduce(
  (result, item) => ({
    ...result,
    [item[key]]: [
      ...(result[item[key]] || []),
      item,
    ],
  }), 
  {},
);

ES2024 native Object.groupBy and Map.groupBy

Object.groupBy([1, 2, 3, 4, 5, 6, 7, 8, 9], v => (v % 2 ? "odd" : "even"));
// { odd: [1, 3, 5, 7, 9], even: [2, 4, 6, 8] };

Map.groupBy([1, 2, 3, 4, 5, 6, 7, 8, 9], v => (v % 2 ? "odd" : "even")).get('odd');
// [1, 3, 5, 7, 9]

GroupBy one-liner, an ES2021 solution

const groupBy = (x,f)=>x.reduce((a,b,i)=>((a[f(b,i,x)]||=[]).push(b),a),{});

TypeScript

const groupBy = <T>(array: T[], predicate: (value: T, index: number, array: T[]) => string) =>
  array.reduce((acc, value, index, array) => {
    (acc[predicate(value, index, array)] ||= []).push(value);
    return acc;
  }, {} as { [key: string]: T[] });

EXAMPLES

const groupBy = (x,f)=>x.reduce((a,b,i)=>((a[f(b,i,x)]||=[]).push(b),a),{});
// f -> should must return string/number because it will be use as key in object

// for demo

groupBy([1, 2, 3, 4, 5, 6, 7, 8, 9], v => (v % 2 ? "odd" : "even"));
// { odd: [1, 3, 5, 7, 9], even: [2, 4, 6, 8] };

const colors = [
  "Apricot",
  "Brown",
  "Burgundy",
  "Cerulean",
  "Peach",
  "Pear",
  "Red",
];

groupBy(colors, v => v[0]); // group by colors name first letter
// {
//   A: ["Apricot"],
//   B: ["Brown", "Burgundy"],
//   C: ["Cerulean"],
//   P: ["Peach", "Pear"],
//   R: ["Red"],
// };

groupBy(colors, v => v.length); // group by length of color names
// {
//   3: ["Red"],
//   4: ["Pear"],
//   5: ["Brown", "Peach"],
//   7: ["Apricot"],
//   8: ["Burgundy", "Cerulean"],
// }

const data = [
  { comment: "abc", forItem: 1, inModule: 1 },
  { comment: "pqr", forItem: 1, inModule: 1 },
  { comment: "klm", forItem: 1, inModule: 2 },
  { comment: "xyz", forItem: 1, inModule: 2 },
];

groupBy(data, v => v.inModule); // group by module
// {
//   1: [
//     { comment: "abc", forItem: 1, inModule: 1 },
//     { comment: "pqr", forItem: 1, inModule: 1 },
//   ],
//   2: [
//     { comment: "klm", forItem: 1, inModule: 2 },
//     { comment: "xyz", forItem: 1, inModule: 2 },
//   ],
// }

groupBy(data, x => x.forItem + "-" + x.inModule); // group by module with item
// {
//   "1-1": [
//     { comment: "abc", forItem: 1, inModule: 1 },
//     { comment: "pqr", forItem: 1, inModule: 1 },
//   ],
//   "1-2": [
//     { comment: "klm", forItem: 1, inModule: 2 },
//     { comment: "xyz", forItem: 1, inModule: 2 },
//   ],
// }

groupByToMap

const groupByToMap = (x, f) =>
  x.reduce((a, b, i, x) => {
    const k = f(b, i, x);
    a.get(k)?.push(b) ?? a.set(k, [b]);
    return a;
  }, new Map());

TypeScript

const groupByToMap = <T, Q>(array: T[], predicate: (value: T, index: number, array: T[]) => Q) =>
  array.reduce((map, value, index, array) => {
    const key = predicate(value, index, array);
    map.get(key)?.push(value) ?? map.set(key, [value]);
    return map;
  }, new Map<Q, T[]>());

You can build an ES6 Map from array.reduce().

const groupedMap = initialArray.reduce(
    (entryMap, e) => entryMap.set(e.id, [...entryMap.get(e.id)||[], e]),
    new Map()
);

This has a few advantages over the other solutions:

• It doesn't require any libraries (unlike e.g. _.groupBy())
• You get a JavaScript Map rather than an object (e.g. as returned by _.groupBy()). This has lots of benefits, including:
  • it remembers the order in which items were first added,
  • keys can be any type rather than just strings.
• A Map is a more useful result that an array of arrays. But if you do want an array of arrays, you can then call Array.from(groupedMap.entries()) (for an array of [key, group array] pairs) or Array.from(groupedMap.values()) (for a simple array of arrays).
• It's quite flexible; often, whatever you were planning to do next with this map can be done directly as part of the reduction.

As an example of the last point, imagine I have an array of objects that I want to do a (shallow) merge on by id, like this:

const objsToMerge = [{id: 1, name: "Steve"}, {id: 2, name: "Alice"}, {id: 1, age: 20}];
// The following variable should be created automatically
const mergedArray = [{id: 1, name: "Steve", age: 20}, {id: 2, name: "Alice"}]

To do this, I would usually start by grouping by id, and then merging each of the resulting arrays. Instead, you can do the merge directly in the reduce():

const mergedArray = Array.from(
    objsToMerge.reduce(
        (entryMap, e) => entryMap.set(e.id, {...entryMap.get(e.id)||{}, ...e}),
        new Map()
    ).values()
);

Later edit:

The above is probably efficient enough for most purposes. But the original question was "most efficient" and, as a couple of people have pointed out, that's not true of the above solution. The problem is mainly that instantiates a new array for every entry. I had thought that this would be optimised away by the JS interpreter but it seems maybe not.

Someone suggested an edit to fix this but it really looked more complicated. Already the original snippet is pushing readability a little. If you really want to do this, please just use a for loop! It's not a sin! It takes one or two more lines of code but it's simpler than functional techniques even though it's not shorter:

const groupedMap = new Map();
for (const e of initialArray) {
    let thisList = groupedMap.get(e.type);
    if (thisList === undefined) {
        thisList = [];
        groupedMap.set(e.type, thisList);
    }
    thisList.push(e);
}

[Edit: updated to even more efficient implementation that avoids doing both .has() and .get() for keys that are already present.]

I would check lodash groupBy it seems to do exactly what you are looking for. It is also quite lightweight and really simple.

Fiddle example: https://jsfiddle.net/r7szvt5k/

Provided that your array name is arr the groupBy with lodash is just:

import groupBy from 'lodash/groupBy';
// if you still use require:
// const groupBy = require('lodash/groupBy');

const a = groupBy(arr, function(n) {
  return n.Phase;
});
// a is your array grouped by Phase attribute

Although the linq answer is interesting, it's also quite heavy-weight. My approach is somewhat different:

var DataGrouper = (function() {
    var has = function(obj, target) {
        return _.any(obj, function(value) {
            return _.isEqual(value, target);
        });
    };

    var keys = function(data, names) {
        return _.reduce(data, function(memo, item) {
            var key = _.pick(item, names);
            if (!has(memo, key)) {
                memo.push(key);
            }
            return memo;
        }, []);
    };

    var group = function(data, names) {
        var stems = keys(data, names);
        return _.map(stems, function(stem) {
            return {
                key: stem,
                vals:_.map(_.where(data, stem), function(item) {
                    return _.omit(item, names);
                })
            };
        });
    };

    group.register = function(name, converter) {
        return group[name] = function(data, names) {
            return _.map(group(data, names), converter);
        };
    };

    return group;
}());

DataGrouper.register("sum", function(item) {
    return _.extend({}, item.key, {Value: _.reduce(item.vals, function(memo, node) {
        return memo + Number(node.Value);
    }, 0)});
});

You can see it in action on JSBin.

I didn't see anything in Underscore that does what has does, although I might be missing it. It's much the same as _.contains, but uses _.isEqual rather than === for comparisons. Other than that, the rest of this is problem-specific, although with an attempt to be generic.

Now DataGrouper.sum(data, ["Phase"]) returns

[
    {Phase: "Phase 1", Value: 50},
    {Phase: "Phase 2", Value: 130}
]

And DataGrouper.sum(data, ["Phase", "Step"]) returns

[
    {Phase: "Phase 1", Step: "Step 1", Value: 15},
    {Phase: "Phase 1", Step: "Step 2", Value: 35},
    {Phase: "Phase 2", Step: "Step 1", Value: 55},
    {Phase: "Phase 2", Step: "Step 2", Value: 75}
]

But sum is only one potential function here. You can register others as you like:

DataGrouper.register("max", function(item) {
    return _.extend({}, item.key, {Max: _.reduce(item.vals, function(memo, node) {
        return Math.max(memo, Number(node.Value));
    }, Number.NEGATIVE_INFINITY)});
});

and now DataGrouper.max(data, ["Phase", "Step"]) will return

[
    {Phase: "Phase 1", Step: "Step 1", Max: 10},
    {Phase: "Phase 1", Step: "Step 2", Max: 20},
    {Phase: "Phase 2", Step: "Step 1", Max: 30},
    {Phase: "Phase 2", Step: "Step 2", Max: 40}
]

or if you registered this:

DataGrouper.register("tasks", function(item) {
    return _.extend({}, item.key, {Tasks: _.map(item.vals, function(item) {
      return item.Task + " (" + item.Value + ")";
    }).join(", ")});
});

then calling DataGrouper.tasks(data, ["Phase", "Step"]) will get you

[
    {Phase: "Phase 1", Step: "Step 1", Tasks: "Task 1 (5), Task 2 (10)"},
    {Phase: "Phase 1", Step: "Step 2", Tasks: "Task 1 (15), Task 2 (20)"},
    {Phase: "Phase 2", Step: "Step 1", Tasks: "Task 1 (25), Task 2 (30)"},
    {Phase: "Phase 2", Step: "Step 2", Tasks: "Task 1 (35), Task 2 (40)"}
]

DataGrouper itself is a function. You can call it with your data and a list of the properties you want to group by. It returns an array whose elements are object with two properties: key is the collection of grouped properties, vals is an array of objects containing the remaining properties not in the key. For example, DataGrouper(data, ["Phase", "Step"]) will yield:

[
    {
        "key": {Phase: "Phase 1", Step: "Step 1"},
        "vals": [
            {Task: "Task 1", Value: "5"},
            {Task: "Task 2", Value: "10"}
        ]
    },
    {
        "key": {Phase: "Phase 1", Step: "Step 2"},
        "vals": [
            {Task: "Task 1", Value: "15"}, 
            {Task: "Task 2", Value: "20"}
        ]
    },
    {
        "key": {Phase: "Phase 2", Step: "Step 1"},
        "vals": [
            {Task: "Task 1", Value: "25"},
            {Task: "Task 2", Value: "30"}
        ]
    },
    {
        "key": {Phase: "Phase 2", Step: "Step 2"},
        "vals": [
            {Task: "Task 1", Value: "35"}, 
            {Task: "Task 2", Value: "40"}
        ]
    }
]

DataGrouper.register accepts a function and creates a new function which accepts the initial data and the properties to group by. This new function then takes the output format as above and runs your function against each of them in turn, returning a new array. The function that's generated is stored as a property of DataGrouper according to a name you supply and also returned if you just want a local reference.

Well that's a lot of explanation. The code is reasonably straightforward, I hope!

This is probably more easily done with linq.js, which is intended to be a true implementation of LINQ in JavaScript (DEMO):

var linq = Enumerable.From(data);
var result =
    linq.GroupBy(function(x){ return x.Phase; })
        .Select(function(x){
          return {
            Phase: x.Key(),
            Value: x.Sum(function(y){ return y.Value|0; })
          };
        }).ToArray();

result:

[
    { Phase: "Phase 1", Value: 50 },
    { Phase: "Phase 2", Value: 130 }
]

Or, more simply using the string-based selectors (DEMO):

linq.GroupBy("$.Phase", "",
    "k,e => { Phase:k, Value:e.Sum('$.Value|0') }").ToArray();

MDN has this example in their Array.reduce() documentation.

// Grouping objects by a property
// https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Array/Reduce#Grouping_objects_by_a_property#Grouping_objects_by_a_property

var people = [
  { name: 'Alice', age: 21 },
  { name: 'Max', age: 20 },
  { name: 'Jane', age: 20 }
];

function groupBy(objectArray, property) {
  return objectArray.reduce(function (acc, obj) {
    var key = obj[property];
    if (!acc[key]) {
      acc[key] = [];
    }
    acc[key].push(obj);
    return acc;
  }, {});
}

var groupedPeople = groupBy(people, 'age');
// groupedPeople is:
// { 
//   20: [
//     { name: 'Max', age: 20 }, 
//     { name: 'Jane', age: 20 }
//   ], 
//   21: [{ name: 'Alice', age: 21 }] 
// }

it's a bit late but maybe someone like this one.

ES6:

const users = [{
    name: "Jim",
    color: "blue"
  },
  {
    name: "Sam",
    color: "blue"
  },
  {
    name: "Eddie",
    color: "green"
  },
  {
    name: "Robert",
    color: "green"
  },
];
const groupBy = (arr, key) => {
  const initialValue = {};
  return arr.reduce((acc, cval) => {
    const myAttribute = cval[key];
    acc[myAttribute] = [...(acc[myAttribute] || []), cval]
    return acc;
  }, initialValue);
};

const res = groupBy(users, "color");
console.log("group by:", res);

_.groupBy([{tipo: 'A' },{tipo: 'A'}, {tipo: 'B'}], 'tipo');
>> Object {A: Array[2], B: Array[1]}

From: http://underscorejs.org/#groupBy

Array.prototype.groupBy = function(keyFunction) {
    var groups = {};
    this.forEach(function(el) {
        var key = keyFunction(el);
        if (key in groups == false) {
            groups[key] = [];
        }
        groups[key].push(el);
    });
    return Object.keys(groups).map(function(key) {
        return {
            key: key,
            values: groups[key]
        };
    });
};

You can do it with Alasql JavaScript library:

var data = [ { Phase: "Phase 1", Step: "Step 1", Task: "Task 1", Value: "5" },
             { Phase: "Phase 1", Step: "Step 1", Task: "Task 2", Value: "10" }];

var res = alasql('SELECT Phase, Step, SUM(CAST([Value] AS INT)) AS [Value] \
                  FROM ? GROUP BY Phase, Step',[data]);

Try this example at jsFiddle.

BTW: On large arrays (100000 records and more) Alasql faster tham Linq. See test at jsPref.

Comments:

• Here I put Value in square brackets, because VALUE is a keyword in SQL
• I have to use CAST() function to convert string Values to number type.

A newer approach with an object for grouping and two more function to create a key and to get an object with wanted items of grouping and another key for the adding value.

const
    groupBy = (array, groups, valueKey) => {
        const
            getKey = o => groups.map(k => o[k]).join('|'),
            getObject = o => Object.fromEntries([...groups.map(k => [k, o[k]]), [valueKey, 0]]);

        groups = [].concat(groups);

        return Object.values(array.reduce((r, o) => {
            (r[getKey(o)] ??= getObject(o))[valueKey] += +o[valueKey];
            return r;
        }, {}));
    },
    data = [{ Phase: "Phase 1", Step: "Step 1", Task: "Task 1", Value: "5" }, { Phase: "Phase 1", Step: "Step 1", Task: "Task 2", Value: "10" }, { Phase: "Phase 1", Step: "Step 2", Task: "Task 1", Value: "15" }, { Phase: "Phase 1", Step: "Step 2", Task: "Task 2", Value: "20" }, { Phase: "Phase 2", Step: "Step 1", Task: "Task 1", Value: "25" }, { Phase: "Phase 2", Step: "Step 1", Task: "Task 2", Value: "30" }, { Phase: "Phase 2", Step: "Step 2", Task: "Task 1", Value: "35" }, { Phase: "Phase 2", Step: "Step 2", Task: "Task 2", Value: "40" }];

console.log(groupBy(data, 'Phase', 'Value'));
console.log(groupBy(data, ['Phase', 'Step'], 'Value'));

.as-console-wrapper { max-height: 100% !important; top: 0; }

Old approach:

Although the question have some answers and the answers look a bit over complicated, I suggest to use vanilla Javascript for group-by with a nested (if necessary) Map.

function groupBy(array, groups, valueKey) {
    var map = new Map;
    groups = [].concat(groups);
    return array.reduce((r, o) => {
        groups.reduce((m, k, i, { length }) => {
            var child;
            if (m.has(o[k])) return m.get(o[k]);
            if (i + 1 === length) {
                child = Object
                    .assign(...groups.map(k => ({ [k]: o[k] })), { [valueKey]: 0 });
                r.push(child);
            } else {
                child = new Map;
            }
            m.set(o[k], child);
            return child;
        }, map)[valueKey] += +o[valueKey];
        return r;
    }, [])
};

var data = [{ Phase: "Phase 1", Step: "Step 1", Task: "Task 1", Value: "5" }, { Phase: "Phase 1", Step: "Step 1", Task: "Task 2", Value: "10" }, { Phase: "Phase 1", Step: "Step 2", Task: "Task 1", Value: "15" }, { Phase: "Phase 1", Step: "Step 2", Task: "Task 2", Value: "20" }, { Phase: "Phase 2", Step: "Step 1", Task: "Task 1", Value: "25" }, { Phase: "Phase 2", Step: "Step 1", Task: "Task 2", Value: "30" }, { Phase: "Phase 2", Step: "Step 2", Task: "Task 1", Value: "35" }, { Phase: "Phase 2", Step: "Step 2", Task: "Task 2", Value: "40" }];

console.log(groupBy(data, 'Phase', 'Value'));
console.log(groupBy(data, ['Phase', 'Step'], 'Value'));

.as-console-wrapper { max-height: 100% !important; top: 0; }

Checked answer -- just shallow grouping. It's pretty nice to understand reducing. Question also provide the problem of additional aggregate calculations.

Here is a REAL GROUP BY for Array of Objects by some field(s) with 1) calculated key name and 2) complete solution for cascading of grouping by providing the list of the desired keys and converting its unique values to root keys like SQL GROUP BY does.

const inputArray = [ 
    { Phase: "Phase 1", Step: "Step 1", Task: "Task 1", Value: "5" },
    { Phase: "Phase 1", Step: "Step 1", Task: "Task 2", Value: "10" },
    { Phase: "Phase 1", Step: "Step 2", Task: "Task 1", Value: "15" },
    { Phase: "Phase 1", Step: "Step 2", Task: "Task 2", Value: "20" },
    { Phase: "Phase 2", Step: "Step 1", Task: "Task 1", Value: "25" },
    { Phase: "Phase 2", Step: "Step 1", Task: "Task 2", Value: "30" },
    { Phase: "Phase 2", Step: "Step 2", Task: "Task 1", Value: "35" },
    { Phase: "Phase 2", Step: "Step 2", Task: "Task 2", Value: "40" }
];

var outObject = inputArray.reduce(function(a, e) {
  // GROUP BY estimated key (estKey), well, may be a just plain key
  // a -- Accumulator result object
  // e -- sequentally checked Element, the Element that is tested just at this itaration

  // new grouping name may be calculated, but must be based on real value of real field
  let estKey = (e['Phase']); 

  (a[estKey] ? a[estKey] : (a[estKey] = null || [])).push(e);
  return a;
}, {});

console.log(outObject);

Play with estKey -- you may group by more then one field, add additional aggregations, calculations or other processing.

Also you can groups data recursively. For example initially group by Phase, then by Step field and so on. Additionally blow off the fat rest data.

const inputArray = [
{ Phase: "Phase 1", Step: "Step 1", Task: "Task 1", Value: "5" },
{ Phase: "Phase 1", Step: "Step 1", Task: "Task 2", Value: "10" },
{ Phase: "Phase 1", Step: "Step 2", Task: "Task 1", Value: "15" },
{ Phase: "Phase 1", Step: "Step 2", Task: "Task 2", Value: "20" },
{ Phase: "Phase 2", Step: "Step 1", Task: "Task 1", Value: "25" },
{ Phase: "Phase 2", Step: "Step 1", Task: "Task 2", Value: "30" },
{ Phase: "Phase 2", Step: "Step 2", Task: "Task 1", Value: "35" },
{ Phase: "Phase 2", Step: "Step 2", Task: "Task 2", Value: "40" }
  ];

/**
 * Small helper to get SHALLOW copy of obj WITHOUT prop
 */
const rmProp = (obj, prop) => ( (({[prop]:_, ...rest})=>rest)(obj) )

/**
 * Group Array by key. Root keys of a resulting array is value
 * of specified key.
 *
 * @param      {Array}   src     The source array
 * @param      {String}  key     The by key to group by
 * @return     {Object}          Object with grouped objects as values
 */
const grpBy = (src, key) => src.reduce((a, e) => (
  (a[e[key]] = a[e[key]] || []).push(rmProp(e, key)),  a
), {});

/**
 * Collapse array of object if it consists of only object with single value.
 * Replace it by the rest value.
 */
const blowObj = obj => Array.isArray(obj) && obj.length === 1 && Object.values(obj[0]).length === 1 ? Object.values(obj[0])[0] : obj;

/**
 * Recursive grouping with list of keys. `keyList` may be an array
 * of key names or comma separated list of key names whom UNIQUE values will
 * becomes the keys of the resulting object.
 */
const grpByReal = function (src, keyList) {
  const [key, ...rest] = Array.isArray(keyList) ? keyList : String(keyList).trim().split(/\s*,\s*/);
  const res = key ? grpBy(src, key) : [...src];
  if (rest.length) {
for (const k in res) {
  res[k] = grpByReal(res[k], rest)
}
  } else {
for (const k in res) {
  res[k] = blowObj(res[k])
}
  }
  return res;
}

console.log( JSON.stringify( grpByReal(inputArray, 'Phase, Step, Task'), null, 2 ) );

Here's a nasty, hard to read solution using ES6:

export default (arr, key) => 
  arr.reduce(
    (r, v, _, __, k = v[key]) => ((r[k] || (r[k] = [])).push(v), r),
    {}
  );

For those asking how does this even work, here's an explanation:

• In both => you have a free return

• The Array.prototype.reduce function takes up to 4 parameters. That's why a fifth parameter is being added so we can have a cheap variable declaration for the group (k) at the parameter declaration level by using a default value. (yes, this is sorcery)

• If our current group doesn't exist on the previous iteration, we create a new empty array ((r[k] || (r[k] = [])) This will evaluate to the leftmost expression, in other words, an existing array or an empty array, this is why there's an immediate push after that expression, because either way you will get an array.

• When there's a return, the comma , operator will discard the leftmost value, returning the tweaked previous group for this scenario.

An easier to understand version that does the same is:

export default (array, key) => 
  array.reduce((previous, currentItem) => {
    const group = currentItem[key];
    if (!previous[group]) previous[group] = [];
    previous[group].push(currentItem);
    return previous;
  }, {});

Edit:

TS Version:

const groupBy = <T, K extends keyof any>(list: T[], getKey: (item: T) => K) =>
  list.reduce((previous, currentItem) => {
    const group = getKey(currentItem);
    if (!previous[group]) previous[group] = [];
    previous[group].push(currentItem);
    return previous;
  }, {} as Record<K, T[]>);

groupByArray(xs, key) {
    return xs.reduce(function (rv, x) {
        let v = key instanceof Function ? key(x) : x[key];
        let el = rv.find((r) => r && r.key === v);
        if (el) {
            el.values.push(x);
        }
        else {
            rv.push({
                key: v,
                values: [x]
            });
        }
        return rv;
    }, []);
}

This one outputs array.

i'd like to suggest my approach. First, separate grouping and aggregating. Lets declare prototypical "group by" function. It takes another function to produce "hash" string for each array element to group by.

Array.prototype.groupBy = function(hash){
  var _hash = hash ? hash : function(o){return o;};

  var _map = {};
  var put = function(map, key, value){
    if (!map[_hash(key)]) {
        map[_hash(key)] = {};
        map[_hash(key)].group = [];
        map[_hash(key)].key = key;

    }
    map[_hash(key)].group.push(value); 
  }

  this.map(function(obj){
    put(_map, obj, obj);
  });

  return Object.keys(_map).map(function(key){
    return {key: _map[key].key, group: _map[key].group};
  });
}

when grouping is done you can aggregate data how you need, in your case

data.groupBy(function(o){return JSON.stringify({a: o.Phase, b: o.Step});})
    /* aggreagating */
    .map(function(el){ 
         var sum = el.group.reduce(
           function(l,c){
             return l + parseInt(c.Value);
           },
           0
         );
         el.key.Value = sum; 
         return el.key;
    });

in common it works. i have tested this code in chrome console. and feel free to improve and find mistakes ;)

This solution takes any arbitrary function (not a key) so it's more flexible than solutions above, and allows arrow functions, which are similar to lambda expressions used in LINQ:

Array.prototype.groupBy = function (funcProp) {
    return this.reduce(function (acc, val) {
        (acc[funcProp(val)] = acc[funcProp(val)] || []).push(val);
        return acc;
    }, {});
};

NOTE: whether you want to extend Array's prototype is up to you.

Example supported in most browsers:

[{a:1,b:"b"},{a:1,c:"c"},{a:2,d:"d"}].groupBy(function(c){return c.a;})

Example using arrow functions (ES6):

[{a:1,b:"b"},{a:1,c:"c"},{a:2,d:"d"}].groupBy(c=>c.a)

Both examples above return:

{
  "1": [{"a": 1, "b": "b"}, {"a": 1, "c": "c"}],
  "2": [{"a": 2, "d": "d"}]
}

Without mutations:

const groupBy = (xs, key) => xs.reduce((acc, x) => Object.assign({}, acc, {
  [x[key]]: (acc[x[key]] || []).concat(x)
}), {})

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

Imagine that you have something like this:

[{id:1, cat:'sedan'},{id:2, cat:'sport'},{id:3, cat:'sport'},{id:4, cat:'sedan'}]

By doing this: const categories = [...new Set(cars.map((car) => car.cat))]

You will get this: ['sedan','sport']

Explanation: 1. First, we are creating a new Set by passing an array. Because Set only allows unique values, all duplicates will be removed.

2. Now the duplicates are gone, we’re going to convert it back to an array by using the spread operator ...

Set Doc:https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Set Spread OperatorDoc: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Operators/Spread_syntax

You can use native JavaScript group array method (currently in stage 2).

I think solution is much more elegant, compared to reduce, or reaching out for third-party libs such as lodash etc.

const products = [{
    name: "milk",
    type: "dairy"
  },
  {
    name: "cheese",
    type: "dairy"
  },
  {
    name: "beef",
    type: "meat"
  },
  {
    name: "chicken",
    type: "meat"
  }
];

const productsByType = products.group((product) => product.type);

console.log("Grouped products by type: ", productsByType);

<script src="https://cdn.jsdelivr.net/npm/[email protected]/minified.min.js"></script>

Most efficient way to group elements in an array of objects in JavaScript is using built-in method:

Object.groupBy()

const input = [
    { Phase: "Phase 1", Step: "Step 1", Task: "Task 1", Value: "5" },
    { Phase: "Phase 2", Step: "Step 1", Task: "Task 2", Value: "30" },
    { Phase: "Phase 2", Step: "Step 2", Task: "Task 1", Value: "35" },
    { Phase: "Phase 2", Step: "Step 2", Task: "Task 2", Value: "40" }
];

const output = Object.groupBy(input, ({ Phase }) => Phase);

console.log(JSON.stringify(output, null, 4));

Result:

{
    "Phase 1": [
        {
            "Phase": "Phase 1",
            "Step": "Step 1",
            "Task": "Task 1",
            "Value": "5"
        }
    ],
    "Phase 2": [
        {
            "Phase": "Phase 2",
            "Step": "Step 1",
            "Task": "Task 2",
            "Value": "30"
        },
        {
            "Phase": "Phase 2",
            "Step": "Step 2",
            "Task": "Task 1",
            "Value": "35"
        },
        {
            "Phase": "Phase 2",
            "Step": "Step 2",
            "Task": "Task 2",
            "Value": "40"
        }
    ]
}

Note: Object.groupBy() method is now supported in Chrome 117 and started to be implemented by other browsers. Check browser compatibility

Polyfill of Object.groupBy in core-js library.

use groupBy in ECMAScript 2024 (ES15)

Object.groupBy()

Map.groupBy()

const inventory = [
  { name: "asparagus", type: "vegetables", quantity: 5 },
  { name: "bananas", type: "fruit", quantity: 0 },
  { name: "goat", type: "meat", quantity: 23 },
  { name: "cherries", type: "fruit", quantity: 5 },
  { name: "fish", type: "meat", quantity: 22 },
];

const result = Object.groupBy(inventory, ({ type }) => type);

console.log(result);

Based on previous answers

const groupBy = (prop) => (xs) =>
  xs.reduce((rv, x) =>
    Object.assign(rv, {[x[prop]]: [...(rv[x[prop]] || []), x]}), {});

and it's a little nicer to look at with object spread syntax, if your environment supports.

const groupBy = (prop) => (xs) =>
  xs.reduce((acc, x) => ({
    ...acc,
    [ x[ prop ] ]: [...( acc[ x[ prop ] ] || []), x],
  }), {});

Here, our reducer takes the partially-formed return value (starting with an empty object), and returns an object composed of the spread out members of the previous return value, along with a new member whose key is calculated from the current iteree's value at prop and whose value is a list of all values for that prop along with the current value.

Here is a ES6 version that won't break on null members

function groupBy (arr, key) {
  return (arr || []).reduce((acc, x = {}) => ({
    ...acc,
    [x[key]]: [...acc[x[key]] || [], x]
  }), {})
}

I would check declarative-js groupBy it seems to do exactly what you are looking for. It is also:

• very performant (performance benchmark)
• written in typescript so all typpings are included.
• It is not enforcing to use 3rd party array-like objects.

import { Reducers } from 'declarative-js';
import groupBy = Reducers.groupBy;
import Map = Reducers.Map;

const data = [
    { Phase: "Phase 1", Step: "Step 1", Task: "Task 1", Value: "5" },
    { Phase: "Phase 1", Step: "Step 1", Task: "Task 2", Value: "10" },
    { Phase: "Phase 1", Step: "Step 2", Task: "Task 1", Value: "15" },
    { Phase: "Phase 1", Step: "Step 2", Task: "Task 2", Value: "20" },
    { Phase: "Phase 2", Step: "Step 1", Task: "Task 1", Value: "25" },
    { Phase: "Phase 2", Step: "Step 1", Task: "Task 2", Value: "30" },
    { Phase: "Phase 2", Step: "Step 2", Task: "Task 1", Value: "35" },
    { Phase: "Phase 2", Step: "Step 2", Task: "Task 2", Value: "40" }
];

data.reduce(groupBy(element=> element.Step), Map());
data.reduce(groupBy('Step'), Map());

Let's fully answer the original question while reusing code that was already written (i.e., Underscore). You can do much more with Underscore if you combine its >100 functions. The following solution demonstrates this.

Step 1: group the objects in the array by an arbitrary combination of properties. This uses the fact that _.groupBy accepts a function that returns the group of an object. It also uses _.chain, _.pick, _.values, _.join and _.value. Note that _.value is not strictly needed here, because chained values will automatically unwrap when used as a property name. I'm including it to safeguard against confusion in case somebody tries to write similar code in a context where automatic unwrapping does not take place.

// Given an object, return a string naming the group it belongs to.
function category(obj) {
    return _.chain(obj).pick(propertyNames).values().join(' ').value();
}

// Perform the grouping.
const intermediate = _.groupBy(arrayOfObjects, category);

Given the arrayOfObjects in the original question and setting propertyNames to ['Phase', 'Step'], intermediate will get the following value:

{
    "Phase 1 Step 1": [
        { Phase: "Phase 1", Step: "Step 1", Task: "Task 1", Value: "5" },
        { Phase: "Phase 1", Step: "Step 1", Task: "Task 2", Value: "10" }
    ],
    "Phase 1 Step 2": [
        { Phase: "Phase 1", Step: "Step 2", Task: "Task 1", Value: "15" },
        { Phase: "Phase 1", Step: "Step 2", Task: "Task 2", Value: "20" }
    ],
    "Phase 2 Step 1": [
        { Phase: "Phase 2", Step: "Step 1", Task: "Task 1", Value: "25" },
        { Phase: "Phase 2", Step: "Step 1", Task: "Task 2", Value: "30" }
    ],
    "Phase 2 Step 2": [
        { Phase: "Phase 2", Step: "Step 2", Task: "Task 1", Value: "35" },
        { Phase: "Phase 2", Step: "Step 2", Task: "Task 2", Value: "40" }
    ]
}

Step 2: reduce each group to a single flat object and return the results in an array. Besides the functions we have seen before, the following code uses _.pluck, _.first, _.pick, _.extend, _.reduce and _.map. _.first is guaranteed to return an object in this case, because _.groupBy does not produce empty groups. _.value is necessary in this case.

// Sum two numbers, even if they are contained in strings.
const addNumeric = (a, b) => +a + +b;

// Given a `group` of objects, return a flat object with their common
// properties and the sum of the property with name `aggregateProperty`.
function summarize(group) {
    const valuesToSum = _.pluck(group, aggregateProperty);
    return _.chain(group).first().pick(propertyNames).extend({
        [aggregateProperty]: _.reduce(valuesToSum, addNumeric)
    }).value();
}

// Get an array with all the computed aggregates.
const result = _.map(intermediate, summarize);

Given the intermediate that we obtained before and setting aggregateProperty to Value, we get the result that the asker desired:

[
    { Phase: "Phase 1", Step: "Step 1", Value: 15 },
    { Phase: "Phase 1", Step: "Step 2", Value: 35 },
    { Phase: "Phase 2", Step: "Step 1", Value: 55 },
    { Phase: "Phase 2", Step: "Step 2", Value: 75 }
]

We can put this all together in a function that takes arrayOfObjects, propertyNames and aggregateProperty as parameters. Note that arrayOfObjects can actually also be a plain object with string keys, because _.groupBy accepts either. For this reason, I have renamed arrayOfObjects to collection.

function aggregate(collection, propertyNames, aggregateProperty) {
    function category(obj) {
        return _.chain(obj).pick(propertyNames).values().join(' ');
    }
    const addNumeric = (a, b) => +a + +b;
    function summarize(group) {
        const valuesToSum = _.pluck(group, aggregateProperty);
        return _.chain(group).first().pick(propertyNames).extend({
            [aggregateProperty]: _.reduce(valuesToSum, addNumeric)
        }).value();
    }
    return _.chain(collection).groupBy(category).map(summarize).value();
}

aggregate(arrayOfObjects, ['Phase', 'Step'], 'Value') will now give us the same result again.

We can take this a step further and enable the caller to compute any statistic over the values in each group. We can do this and also enable the caller to add arbitrary properties to the summary of each group. We can do all of this while making our code shorter. We replace the aggregateProperty parameter by an iteratee parameter and pass this straight to _.reduce:

function aggregate(collection, propertyNames, iteratee) {
    function category(obj) {
        return _.chain(obj).pick(propertyNames).values().join(' ');
    }
    function summarize(group) {
        return _.chain(group).first().pick(propertyNames)
            .extend(_.reduce(group, iteratee)).value();
    }
    return _.chain(collection).groupBy(category).map(summarize).value();
}

In effect, we move some of the responsibility to the caller; she must provide an iteratee that can be passed to _.reduce, so that the call to _.reduce will produce an object with the aggregate properties she wants to add. For example, we obtain the same result as before with the following expression:

aggregate(arrayOfObjects, ['Phase', 'Step'], (memo, value) => ({
    Value: +memo.Value + +value.Value
}));

For an example of a slightly more sophisticated iteratee, suppose that we want to compute the maximum Value of each group instead of the sum, and that we want to add a Tasks property that lists all the values of Task that occur in the group. Here's one way we can do this, using the last version of aggregate above (and _.union):

aggregate(arrayOfObjects, ['Phase', 'Step'], (memo, value) => ({
    Value: Math.max(memo.Value, value.Value),
    Tasks: _.union(memo.Tasks || [memo.Task], [value.Task])
}));

We obtain the following result:

[
    { Phase: "Phase 1", Step: "Step 1", Value: 10, Tasks: [ "Task 1", "Task 2" ] },
    { Phase: "Phase 1", Step: "Step 2", Value: 20, Tasks: [ "Task 1", "Task 2" ] },
    { Phase: "Phase 2", Step: "Step 1", Value: 30, Tasks: [ "Task 1", "Task 2" ] },
    { Phase: "Phase 2", Step: "Step 2", Value: 40, Tasks: [ "Task 1", "Task 2" ] }
]

Credit to @much2learn, who also posted an answer that can handle arbitrary reducing functions. I wrote a couple more SO answers that demonstrate how one can achieve sophisticated things by combining multiple Underscore functions:

• https://stackoverflow.com/a/64938636/1166087
• https://stackoverflow.com/a/64094738/1166087
• https://stackoverflow.com/a/63625129/1166087
• https://stackoverflow.com/a/63088916/1166087

groupBy function that can group an array by a specific key or a given grouping function. Typed.

groupBy = <T, K extends keyof T>(array: T[], groupOn: K | ((i: T) => string)): Record<string, T[]> => {
  const groupFn = typeof groupOn === 'function' ? groupOn : (o: T) => o[groupOn];

  return Object.fromEntries(
    array.reduce((acc, obj) => {
      const groupKey = groupFn(obj);
      return acc.set(groupKey, [...(acc.get(groupKey) || []), obj]);
    }, new Map())
  ) as Record<string, T[]>;
};