¿Qué uso para una implementación max-heap en Python?

Python incluye el módulo heapq para montones mínimos, pero necesito un montón máximo. ¿Qué debo usar para una implementación max-heap en Python?

La forma más fácil es invertir el valor de las claves y usar heapq. Por ejemplo, convierta 1000.0 en -1000.0 y 5.0 en -5.0.

Puedes usar

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!!

Si luego quieres hacer estallar elementos, usa:

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

La solución es negar sus valores cuando los almacena en el montón, o invertir su comparación de objetos de la siguiente manera:

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)

Ejemplo de un montón máximo:

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

Pero debe recordar envolver y desenvolver sus valores, lo que requiere saber si se trata de un montón mínimo o máximo.

Clases MinHeap, MaxHeap

Agregar clases para MinHeapy MaxHeapobjetos puede simplificar su código:

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

Ejemplo de uso:

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"

La solución más fácil e ideal.

Multiplica los valores por -1

Ahí tienes. Todos los números más altos son ahora los más bajos y viceversa.

Solo recuerde que cuando hace estallar un elemento para multiplicarlo por -1 para obtener el valor original nuevamente.

Implementé una versión heap max de heapq y la envié a PyPI. (Muy ligero cambio del código CPython del módulo heapq.)

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

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

Instalación

pip install heapq_max

Uso

tl; dr: igual que el módulo heapq excepto que agrega '_max' a todas las funciones.

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

Si está insertando claves que son comparables pero no de tipo int, podría anular los operadores de comparación (es decir, <= convertirse> y> se convierte en <=). De lo contrario, puede anular heapq._siftup en el módulo heapq (al final, todo es solo código Python).

Permitiéndole elegir una cantidad arbitraria de artículos más grandes o más pequeños

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]

Ampliar la clase int y anular __lt__ es una de las formas.

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()

He creado un contenedor de montón que invierte los valores para crear un montón máximo, así como una clase de contenedor para un montón mínimo para hacer que la biblioteca sea más parecida a OOP. Aquí está la esencia. Hay tres clases; Heap (clase abstracta), HeapMin y HeapMax.

Métodos:

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

En caso de que desee obtener el elemento K más grande utilizando el montón máximo, puede hacer el siguiente truco:

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]

Siguiendo la excelente respuesta de Isaac Turner , me gustaría poner un ejemplo basado en K Puntos más cercanos al origen usando el montón máximo.

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)

Para profundizar en https://stackoverflow.com/a/59311063/1328979 , aquí hay una implementación de Python 3 completamente documentada, comentada y probada para el caso general.

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

Esta es una MaxHeapimplementación simple basada en heapq. Aunque solo funciona con valores numéricos.

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)

Uso:

max_heap = MaxHeap()
max_heap.push(3)
max_heap.push(5)
max_heap.push(1)
print(max_heap.top())  # 5