Pythonの最大ヒープ実装には何を使用しますか?


回答:


244

最も簡単な方法は、キーの値を反転してheapqを使用することです。たとえば、1000.0を-1000.0に、5.0を-5.0に変更します。


38
これも標準ソリューションです。
Andrew McGregor

44
うーん; 総クラッジ。私はheapq逆を提供しないことに驚いています。
shabbychef 2010

40
ワオ。私はよびっくり、これはによって提供されていないことheapq、そして何の良い代替がないこと。
ire_and_curses

23
@gatoatigrado:int/ floatに簡単にマッピングできないものがある場合は、反転__lt__演算子を使用してクラスにラップすることで順序を逆にすることができます。
Daniel Stutzbach、2012

5
@Aerovistae同じアドバイスが適用されます。最初から正か負かに関係なく、値を反転します(つまり、符号を切り替えます)。
Dennis

235

使用できます

import heapq
listForTree = [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]    
heapq.heapify(listForTree)             # for a min heap
heapq._heapify_max(listForTree)        # for a maxheap!!

その後、要素をポップしたい場合は、次を使用します。

heapq.heappop(minheap)      # pop from minheap
heapq._heappop_max(maxheap) # pop from maxheap

34
:そこのように見えますが、最大ヒープのためのいくつかの文書化されていない機能があり_heapify_max_heappushpop_max_siftdown_max、と_siftup_max
ziyuang 2014

127
ワオ。私はよびっくりあっていることですなAビルトインheapqで解決。しかし、完全にある不合理それがされていることはありません少しでも公式文書に一切言及しました!WTF!
RayLuo 2015

27
pop / push関数はいずれも最大ヒープ構造を破壊するため、この方法は実行できません。
シッダールタ2017

22
使用しないでください。LinMaとSiddharthaが気づいたように、push / popは注文を破棄します。
Alex Fedulov 2017

13
アンダースコアで始まるメソッドはプライベートであり、事前の通知なしに削除できます。使用しないでください。
user4815162342

66

解決策は、ヒープに値を格納するときに値を無効にするか、次のようにオブジェクト比較を反転させることです。

import heapq

class MaxHeapObj(object):
  def __init__(self, val): self.val = val
  def __lt__(self, other): return self.val > other.val
  def __eq__(self, other): return self.val == other.val
  def __str__(self): return str(self.val)

最大ヒープの例:

maxh = []
heapq.heappush(maxh, MaxHeapObj(x))
x = maxh[0].val  # fetch max value
x = heapq.heappop(maxh).val  # pop max value

ただし、値をラップおよびアンラップすることを覚えておく必要があります。これには、最小ヒープと最大ヒープのどちらを扱っているかを知る必要があります。

MinHeap、MaxHeapクラス

MinHeapおよびMaxHeapオブジェクトのクラスを追加すると、コードを簡略化できます。

class MinHeap(object):
  def __init__(self): self.h = []
  def heappush(self, x): heapq.heappush(self.h, x)
  def heappop(self): return heapq.heappop(self.h)
  def __getitem__(self, i): return self.h[i]
  def __len__(self): return len(self.h)

class MaxHeap(MinHeap):
  def heappush(self, x): heapq.heappush(self.h, MaxHeapObj(x))
  def heappop(self): return heapq.heappop(self.h).val
  def __getitem__(self, i): return self.h[i].val

使用例:

minh = MinHeap()
maxh = MaxHeap()
# add some values
minh.heappush(12)
maxh.heappush(12)
minh.heappush(4)
maxh.heappush(4)
# fetch "top" values
print(minh[0], maxh[0])  # "4 12"
# fetch and remove "top" values
print(minh.heappop(), maxh.heappop())  # "4 12"

いいね。私はこれを取りlist、__ init__にオプションのパラメーターをheapq.heapify追加しましたheapreplace。この場合、私は呼び出し、メソッドも追加しました。
Booboo

1
このタイプミスMaxHeapInt-> MaxHeapObjを誰も捕まえなかったことに驚いた。そうでなければ、確かに非常にクリーンなソリューションです。
Chiraz BenAbdelkader

@ChirazBenAbdelkader修正、ありがとうございます。
アイザックターナー

39

最も簡単で理想的なソリューション

値に-1を掛けます

どうぞ。最高の数字はすべて最低となり、その逆も同様です。

元の値を再び取得するために、要素をポップして-1を掛けるときは、覚えておいてください。


すばらしいですが、ほとんどのソリューションはクラス/その他のタイプをサポートしており、実際のデータを変更しません。未解決の問題は、値に-1を乗算しても値が変化しないかどうか(非常に正確な浮動小数点数)です。
Alex Baranowski

1
@AlexBaranowski。それは本当ですが、それはメンテナからの応答でした
Flair

メンテナにはいくつかの機能を実装しない権利がありますが、このIMOは実際に役立ちます。
Alex Baranowski

7

heapqの最大ヒープバージョンを実装し、PyPIに送信しました。(ヒープモジュールCPythonコードの非常にわずかな変更。)

https://pypi.python.org/pypi/heapq_max/

https://github.com/he-zhe/heapq_max

取り付け

pip install heapq_max

使用法

tl; dr:すべての関数に「_max」を追加することを除いて、heapqモジュールと同じです。

heap_max = []                           # creates an empty heap
heappush_max(heap_max, item)            # pushes a new item on the heap
item = heappop_max(heap_max)            # pops the largest item from the heap
item = heap_max[0]                      # largest item on the heap without popping it
heapify_max(x)                          # transforms list into a heap, in-place, in linear time
item = heapreplace_max(heap_max, item)  # pops and returns largest item, and
                                    # adds new item; the heap size is unchanged

4

比較可能であるがintに似ていないキーを挿入する場合、それらの比較演算子をオーバーライドする可能性があります(つまり、<=が>になり、>が<=になります)。それ以外の場合は、heapqモジュールのheapq._siftupをオーバーライドできます(結局のところ、すべてPythonコードのみです)。


9
「すべてPythonコードだけです」:Pythonのバージョンとインストールによって異なります。たとえば、インストールしたheapq.pyには、309行目(# If available, use C implementation)の後にコメントが正確に説明するコードが含まれています。
tzot

3

最大または最小のアイテムの任意の量を選択できるようにします

import heapq
heap = [23, 7, -4, 18, 23, 42, 37, 2, 8, 2, 23, 7, -4, 18, 23, 42, 37, 2]
heapq.heapify(heap)
print(heapq.nlargest(3, heap))  # [42, 42, 37]
print(heapq.nsmallest(3, heap)) # [-4, -4, 2]

3
説明が正しいでしょう。
Peter Mortensen

私のタイトルは私の説明です
jasonleonhard

1
私の答えは質問よりも長いです。どのような説明を追加しますか?
jasonleonhard


2
これにより正しい結果が得られますが、実際にヒープを使用して効率を上げることはありません。このドキュメントでは、nlargestとnsmallestがリストを毎回ソートするように指定しています。
RossFabricant

3

intクラスを拡張して__lt__をオーバーライドするのも1つの方法です。

import queue
class MyInt(int):
    def __lt__(self, other):
        return self > other

def main():
    q = queue.PriorityQueue()
    q.put(MyInt(10))
    q.put(MyInt(5))
    q.put(MyInt(1))
    while not q.empty():
        print (q.get())


if __name__ == "__main__":
    main()

それは可能ですが、それは物事をかなり遅くし、多くの余分なメモリを使用するように感じます。MyIntは、ヒープ構造の外でも実際に使用することはできません。しかし、例を入力していただき、ありがとうございます。
レオUfimtsev

はぁ!コメントした翌日、カスタムオブジェクトをヒープに入れて最大ヒープが必要になる状況に遭遇しました。私は実際にこの投稿を再度ググったところ、あなたの答えが見つかり、それに基づいて私の解決策を見つけました。(カスタムオブジェクトは、x、y座標を持つポイントであり、ltは中心からの距離を比較してオーバーライドします)。これを投稿してくれてありがとう、賛成です!
レオUfimtsev

1

値を反転してmax-heapを作成するヒープラッパーと、min-heapのラッパークラスを作成して、ライブラリをよりOOPのようにしました。ここにに要点があります。クラスは3つあります。ヒープ(抽象クラ​​ス)、HeapMin、およびHeapMax。

方法:

isempty() -> bool; obvious
getroot() -> int; returns min/max
push() -> None; equivalent to heapq.heappush
pop() -> int; equivalent to heapq.heappop
view_min()/view_max() -> int; alias for getroot()
pushpop() -> int; equivalent to heapq.pushpop

0

最大ヒープを使用して最大のK要素を取得したい場合は、次のトリックを実行できます。

nums= [3,2,1,5,6,4]
k = 2  #k being the kth largest element you want to get
heapq.heapify(nums) 
temp = heapq.nlargest(k, nums)
return temp[-1]

1
残念ながら、これの時間の複雑さはO(MlogM)であり、M = len(nums)であり、これはheapqの目的に反します。実装とコメントを参照してくださいnlargestここを- > github.com/python/cpython/blob/...
アーサー・S

1
有益なコメントをありがとうございます、添付のリンクを必ず確認してください。
RowanX

0

アイザックターナーの素晴らしい答えに続き、最大ヒープを使用して、原点に最も近いK点に基づく例を示します。

from math import sqrt
import heapq


class MaxHeapObj(object):
    def __init__(self, val):
        self.val = val.distance
        self.coordinates = val.coordinates

    def __lt__(self, other):
        return self.val > other.val

    def __eq__(self, other):
        return self.val == other.val

    def __str__(self):
        return str(self.val)


class MinHeap(object):
    def __init__(self):
        self.h = []

    def heappush(self, x):
        heapq.heappush(self.h, x)

    def heappop(self):
        return heapq.heappop(self.h)

    def __getitem__(self, i):
        return self.h[i]

    def __len__(self):
        return len(self.h)


class MaxHeap(MinHeap):
    def heappush(self, x):
        heapq.heappush(self.h, MaxHeapObj(x))

    def heappop(self):
        return heapq.heappop(self.h).val

    def peek(self):
        return heapq.nsmallest(1, self.h)[0].val

    def __getitem__(self, i):
        return self.h[i].val


class Point():
    def __init__(self, x, y):
        self.distance = round(sqrt(x**2 + y**2), 3)
        self.coordinates = (x, y)


def find_k_closest(points, k):
    res = [Point(x, y) for (x, y) in points]
    maxh = MaxHeap()

    for i in range(k):
        maxh.heappush(res[i])

    for p in res[k:]:
        if p.distance < maxh.peek():
            maxh.heappop()
            maxh.heappush(p)

    res = [str(x.coordinates) for x in maxh.h]
    print(f"{k} closest points from origin : {', '.join(res)}")


points = [(10, 8), (-2, 4), (0, -2), (-1, 0), (3, 5), (-2, 3), (3, 2), (0, 1)]
find_k_closest(points, 3)

0

https://stackoverflow.com/a/59311063/1328979について詳しく説明するために、完全に文書化され、注釈が付けられ、テストされた一般的なケースのPython 3実装を示します。

from __future__ import annotations  # To allow "MinHeap.push -> MinHeap:"
from typing import Generic, List, Optional, TypeVar
from heapq import heapify, heappop, heappush, heapreplace


T = TypeVar('T')


class MinHeap(Generic[T]):
    '''
    MinHeap provides a nicer API around heapq's functionality.
    As it is a minimum heap, the first element of the heap is always the
    smallest.
    >>> h = MinHeap([3, 1, 4, 2])
    >>> h[0]
    1
    >>> h.peek()
    1
    >>> h.push(5)  # N.B.: the array isn't always fully sorted.
    [1, 2, 4, 3, 5]
    >>> h.pop()
    1
    >>> h.pop()
    2
    >>> h.pop()
    3
    >>> h.push(3).push(2)
    [2, 3, 4, 5]
    >>> h.replace(1)
    2
    >>> h
    [1, 3, 4, 5]
    '''
    def __init__(self, array: Optional[List[T]] = None):
        if array is None:
            array = []
        heapify(array)
        self.h = array
    def push(self, x: T) -> MinHeap:
        heappush(self.h, x)
        return self  # To allow chaining operations.
    def peek(self) -> T:
        return self.h[0]
    def pop(self) -> T:
        return heappop(self.h)
    def replace(self, x: T) -> T:
        return heapreplace(self.h, x)
    def __getitem__(self, i) -> T:
        return self.h[i]
    def __len__(self) -> int:
        return len(self.h)
    def __str__(self) -> str:
        return str(self.h)
    def __repr__(self) -> str:
        return str(self.h)


class Reverse(Generic[T]):
    '''
    Wrap around the provided object, reversing the comparison operators.
    >>> 1 < 2
    True
    >>> Reverse(1) < Reverse(2)
    False
    >>> Reverse(2) < Reverse(1)
    True
    >>> Reverse(1) <= Reverse(2)
    False
    >>> Reverse(2) <= Reverse(1)
    True
    >>> Reverse(2) <= Reverse(2)
    True
    >>> Reverse(1) == Reverse(1)
    True
    >>> Reverse(2) > Reverse(1)
    False
    >>> Reverse(1) > Reverse(2)
    True
    >>> Reverse(2) >= Reverse(1)
    False
    >>> Reverse(1) >= Reverse(2)
    True
    >>> Reverse(1)
    1
    '''
    def __init__(self, x: T) -> None:
        self.x = x
    def __lt__(self, other: Reverse) -> bool:
        return other.x.__lt__(self.x)
    def __le__(self, other: Reverse) -> bool:
        return other.x.__le__(self.x)
    def __eq__(self, other) -> bool:
        return self.x == other.x
    def __ne__(self, other: Reverse) -> bool:
        return other.x.__ne__(self.x)
    def __ge__(self, other: Reverse) -> bool:
        return other.x.__ge__(self.x)
    def __gt__(self, other: Reverse) -> bool:
        return other.x.__gt__(self.x)
    def __str__(self):
        return str(self.x)
    def __repr__(self):
        return str(self.x)


class MaxHeap(MinHeap):
    '''
    MaxHeap provides an implement of a maximum-heap, as heapq does not provide
    it. As it is a maximum heap, the first element of the heap is always the
    largest. It achieves this by wrapping around elements with Reverse,
    which reverses the comparison operations used by heapq.
    >>> h = MaxHeap([3, 1, 4, 2])
    >>> h[0]
    4
    >>> h.peek()
    4
    >>> h.push(5)  # N.B.: the array isn't always fully sorted.
    [5, 4, 3, 1, 2]
    >>> h.pop()
    5
    >>> h.pop()
    4
    >>> h.pop()
    3
    >>> h.pop()
    2
    >>> h.push(3).push(2).push(4)
    [4, 3, 2, 1]
    >>> h.replace(1)
    4
    >>> h
    [3, 1, 2, 1]
    '''
    def __init__(self, array: Optional[List[T]] = None):
        if array is not None:
            array = [Reverse(x) for x in array]  # Wrap with Reverse.
        super().__init__(array)
    def push(self, x: T) -> MaxHeap:
        super().push(Reverse(x))
        return self
    def peek(self) -> T:
        return super().peek().x
    def pop(self) -> T:
        return super().pop().x
    def replace(self, x: T) -> T:
        return super().replace(Reverse(x)).x


if __name__ == '__main__':
    import doctest
    doctest.testmod()

https://gist.github.com/marccarre/577a55850998da02af3d4b7b98152cf4


0

これはにMaxHeap基づく単純な実装ですheapq。ただし、数値でのみ機能します。

import heapq
from typing import List


class MaxHeap:
    def __init__(self):
        self.data = []

    def top(self):
        return -self.data[0]

    def push(self, val):
        heapq.heappush(self.data, -val)

    def pop(self):
        return -heapq.heappop(self.data)

使用法:

max_heap = MaxHeap()
max_heap.push(3)
max_heap.push(5)
max_heap.push(1)
print(max_heap.top())  # 5
弊社のサイトを使用することにより、あなたは弊社のクッキーポリシーおよびプライバシーポリシーを読み、理解したものとみなされます。
Licensed under cc by-sa 3.0 with attribution required.