mirror of
https://github.com/TheAlgorithms/Python.git
synced 2024-11-30 16:31:08 +00:00
Compare commits
4 Commits
9007583665
...
d20f6c756d
Author | SHA1 | Date | |
---|---|---|---|
|
d20f6c756d | ||
|
071ce716e4 | ||
|
a92936bd01 | ||
|
ea5a187b0a |
|
@ -1,444 +1,361 @@
|
|||
from __future__ import annotations
|
||||
"""
|
||||
Fibonacci Heap
|
||||
A more efficient priority queue implementation that provides amortized time bounds
|
||||
that are better than those of the binary and binomial heaps.
|
||||
reference: https://en.wikipedia.org/wiki/Fibonacci_heap
|
||||
|
||||
Operations supported:
|
||||
- Insert: O(1) amortized
|
||||
- Find minimum: O(1)
|
||||
- Delete minimum: O(log n) amortized
|
||||
- Decrease key: O(1) amortized
|
||||
- Merge: O(1)
|
||||
"""
|
||||
|
||||
|
||||
class Node:
|
||||
"""A node in the Fibonacci heap.
|
||||
"""
|
||||
A node in a Fibonacci heap.
|
||||
|
||||
Each node maintains references to its key, degree (number of children),
|
||||
marked status, parent, child, and circular linked list references (left/right).
|
||||
Args:
|
||||
val: The value stored in the node.
|
||||
|
||||
Attributes:
|
||||
key: The key value stored in the node
|
||||
degree: Number of children of the node
|
||||
marked: Boolean indicating if the node is marked
|
||||
parent: Reference to parent node
|
||||
child: Reference to one child node
|
||||
left: Reference to left sibling in circular list
|
||||
right: Reference to right sibling in circular list
|
||||
|
||||
Examples:
|
||||
>>> node = Node(5)
|
||||
>>> node.key
|
||||
5
|
||||
>>> node.degree
|
||||
0
|
||||
>>> node.marked
|
||||
False
|
||||
>>> node.left == node
|
||||
True
|
||||
>>> node.right == node
|
||||
True
|
||||
val: The value stored in the node.
|
||||
parent: Reference to parent node.
|
||||
child: Reference to one child node.
|
||||
left: Reference to left sibling.
|
||||
right: Reference to right sibling.
|
||||
degree: Number of children.
|
||||
mark: Boolean indicating if node has lost a child.
|
||||
"""
|
||||
|
||||
def __init__(self, key: float | None) -> None:
|
||||
self.key = key or None
|
||||
self.degree = 0
|
||||
self.marked = False
|
||||
self.parent = Node(None)
|
||||
self.child = Node(None)
|
||||
def __init__(self, val):
|
||||
self.val = val
|
||||
self.parent = None
|
||||
self.child = None
|
||||
self.left = self
|
||||
self.right = self
|
||||
self.degree = 0
|
||||
self.mark = False
|
||||
|
||||
def add_sibling(self, node):
|
||||
"""
|
||||
Adds a node as a sibling to the current node.
|
||||
|
||||
Args:
|
||||
node: The node to add as a sibling.
|
||||
"""
|
||||
node.left = self
|
||||
node.right = self.right
|
||||
self.right.left = node
|
||||
self.right = node
|
||||
|
||||
def add_child(self, node):
|
||||
"""
|
||||
Adds a node as a child of the current node.
|
||||
|
||||
Args:
|
||||
node: The node to add as a child.
|
||||
"""
|
||||
node.parent = self
|
||||
if not self.child:
|
||||
self.child = node
|
||||
else:
|
||||
self.child.add_sibling(node)
|
||||
self.degree += 1
|
||||
|
||||
def remove(self):
|
||||
"""Removes this node from its sibling list."""
|
||||
self.left.right = self.right
|
||||
self.right.left = self.left
|
||||
|
||||
|
||||
class FibonacciHeap:
|
||||
"""Implementation of a Fibonacci heap using circular linked lists.
|
||||
|
||||
A Fibonacci heap is a collection of trees satisfying the min-heap property.
|
||||
This implementation uses circular linked lists for both the root list and
|
||||
child lists of nodes.
|
||||
|
||||
Attributes:
|
||||
min_node: Reference to the node with minimum key
|
||||
total_nodes: Total number of nodes in the heap
|
||||
|
||||
Reference: Introduction to Algorithms (CLRS) Chapter 19
|
||||
https://en.wikipedia.org/wiki/Fibonacci_heap
|
||||
|
||||
Examples:
|
||||
>>> heap = FibonacciHeap()
|
||||
>>> heap.is_empty()
|
||||
True
|
||||
>>> node = heap.insert(3)
|
||||
>>> node.key
|
||||
3
|
||||
>>> node2 = heap.insert(2)
|
||||
>>> node2.key
|
||||
2
|
||||
>>> heap.find_min()
|
||||
2
|
||||
>>> heap.extract_min()
|
||||
2
|
||||
>>> heap.find_min()
|
||||
3
|
||||
"""
|
||||
A Fibonacci heap implementation providing
|
||||
amortized efficient priority queue operations.
|
||||
|
||||
def __init__(self) -> None:
|
||||
self.min_node = Node(None)
|
||||
self.total_nodes = 0
|
||||
|
||||
def insert(self, key: float | None) -> Node:
|
||||
"""Insert a new key into the heap.
|
||||
|
||||
Args:
|
||||
key: The key value to insert
|
||||
|
||||
Returns:
|
||||
Node: The newly created node
|
||||
|
||||
Examples:
|
||||
>>> heap = FibonacciHeap()
|
||||
>>> node = heap.insert(5)
|
||||
>>> node.key
|
||||
5
|
||||
>>> heap.find_min()
|
||||
5
|
||||
>>> node2 = heap.insert(3)
|
||||
>>> node2.key
|
||||
3
|
||||
>>> heap.find_min()
|
||||
3
|
||||
"""
|
||||
new_node = Node(key)
|
||||
if self.min_node is None:
|
||||
self.min_node = new_node
|
||||
else:
|
||||
self._insert_into_circular_list(self.min_node, new_node)
|
||||
if new_node.key < self.min_node.key:
|
||||
self.min_node = new_node
|
||||
self.total_nodes += 1
|
||||
return new_node
|
||||
|
||||
@staticmethod
|
||||
def _insert_into_circular_list(base_node: Node, node_to_insert: Node) -> Node:
|
||||
"""Insert node into circular linked list.
|
||||
|
||||
Args:
|
||||
base_node: The reference node in the circular list
|
||||
node_to_insert: The node to insert into the list
|
||||
|
||||
Returns:
|
||||
Node: The base node
|
||||
|
||||
Examples:
|
||||
>>> heap = FibonacciHeap()
|
||||
>>> node1 = Node(1)
|
||||
>>> node2 = Node(2)
|
||||
>>> result = heap._insert_into_circular_list(node1, node2)
|
||||
>>> result == node1
|
||||
True
|
||||
>>> node1.right == node2
|
||||
True
|
||||
>>> node2.left == node1
|
||||
True
|
||||
"""
|
||||
if base_node.key is None:
|
||||
return node_to_insert
|
||||
|
||||
node_to_insert.right = base_node.right
|
||||
node_to_insert.left = base_node
|
||||
base_node.right.left = node_to_insert
|
||||
base_node.right = node_to_insert
|
||||
return base_node
|
||||
|
||||
def extract_min(self) -> float | None:
|
||||
"""Remove and return the minimum key from the heap.
|
||||
|
||||
This operation removes the node with the minimum key from the heap,
|
||||
adds all its children to the root list, and consolidates the heap
|
||||
to maintain the Fibonacci heap properties. This is one of the more
|
||||
complex operations with amortized time complexity of O(log n).
|
||||
|
||||
Returns:
|
||||
Node: The minimum key value that was removed,
|
||||
or None if the heap is empty
|
||||
This implementation provides the following time complexities:
|
||||
- Insert: O(1) amortized
|
||||
- Find minimum: O(1)
|
||||
- Delete minimum: O(log n) amortized
|
||||
- Decrease key: O(1) amortized
|
||||
- Merge: O(1)
|
||||
|
||||
Example:
|
||||
>>> heap = FibonacciHeap()
|
||||
>>> node1 = heap.insert(3)
|
||||
>>> node2 = heap.insert(1)
|
||||
>>> node3 = heap.insert(2)
|
||||
>>> heap.extract_min() # Removes and returns 1
|
||||
1
|
||||
>>> heap.extract_min() # Removes and returns 2
|
||||
>>> node2 = heap.insert(2)
|
||||
>>> node3 = heap.insert(15)
|
||||
>>> heap.peek()
|
||||
2
|
||||
>>> heap.extract_min() # Removes and returns 3
|
||||
>>> heap.delete_min()
|
||||
2
|
||||
>>> heap.peek()
|
||||
3
|
||||
>>> heap.extract_min() # Heap is now empty
|
||||
|
||||
Note:
|
||||
This operation may trigger heap consolidation to maintain
|
||||
the Fibonacci heap properties after removal of the minimum node.
|
||||
>>> other_heap = FibonacciHeap()
|
||||
>>> node4 = other_heap.insert(1)
|
||||
>>> heap.merge_heaps(other_heap)
|
||||
>>> heap.peek()
|
||||
1
|
||||
"""
|
||||
if self.min_node is None:
|
||||
return Node(None).key
|
||||
|
||||
min_node = self.min_node
|
||||
def __init__(self):
|
||||
"""Initializes an empty Fibonacci heap."""
|
||||
self.min_node = None
|
||||
self.size = 0
|
||||
|
||||
if min_node.child:
|
||||
current_child = min_node.child
|
||||
last_child = min_node.child.left
|
||||
while True:
|
||||
next_child = current_child.right
|
||||
self._insert_into_circular_list(self.min_node, current_child)
|
||||
current_child.parent.key = None
|
||||
if current_child == last_child:
|
||||
break
|
||||
current_child = next_child
|
||||
def is_empty(self):
|
||||
"""
|
||||
Checks if the heap is empty.
|
||||
|
||||
min_node.left.right = min_node.right
|
||||
min_node.right.left = min_node.left
|
||||
Returns:
|
||||
bool: True if heap is empty, False otherwise.
|
||||
"""
|
||||
return self.min_node is None
|
||||
|
||||
if min_node == min_node.right:
|
||||
self.min_node.key = None
|
||||
def insert(self, val):
|
||||
"""
|
||||
Inserts a new value into the heap.
|
||||
|
||||
Args:
|
||||
val: Value to insert.
|
||||
|
||||
Returns:
|
||||
Node: The newly created node.
|
||||
"""
|
||||
node = Node(val)
|
||||
if not self.min_node:
|
||||
self.min_node = node
|
||||
else:
|
||||
self.min_node = min_node.right
|
||||
self._consolidate()
|
||||
self.min_node.add_sibling(node)
|
||||
if node.val < self.min_node.val:
|
||||
self.min_node = node
|
||||
self.size += 1
|
||||
return node
|
||||
|
||||
self.total_nodes -= 1
|
||||
return min_node.key
|
||||
|
||||
def _consolidate(self) -> None:
|
||||
"""Consolidate the heap after removing the minimum node.
|
||||
|
||||
This internal method maintains the Fibonacci heap properties by combining
|
||||
trees of the same degree until no two roots have the same degree. This
|
||||
process is key to maintaining the efficiency of the data structure.
|
||||
|
||||
The consolidation process works by:
|
||||
1. Creating a temporary array indexed by tree degree
|
||||
2. Processing each root node and combining trees of the same degree
|
||||
3. Reconstructing the root list and finding the new minimum
|
||||
|
||||
Time complexity: O(log n) amortized
|
||||
|
||||
Note:
|
||||
This is an internal method called by extract_min and should not be
|
||||
called directly from outside the class.
|
||||
def peek(self):
|
||||
"""
|
||||
max_degree = int(self.total_nodes**0.5) + 1
|
||||
degree_table = [Node(None)] * max_degree
|
||||
Returns the minimum value without removing it.
|
||||
|
||||
Returns:
|
||||
The minimum value in the heap.
|
||||
|
||||
Raises:
|
||||
IndexError: If the heap is empty.
|
||||
"""
|
||||
if not self.min_node:
|
||||
raise IndexError("Heap is empty")
|
||||
return self.min_node.val
|
||||
|
||||
def merge_heaps(self, other):
|
||||
"""
|
||||
Merges another Fibonacci heap into this one.
|
||||
|
||||
Args:
|
||||
other: Another FibonacciHeap instance to merge with this one.
|
||||
"""
|
||||
if not other.min_node:
|
||||
return
|
||||
if not self.min_node:
|
||||
self.min_node = other.min_node
|
||||
else:
|
||||
# Merge root lists
|
||||
self.min_node.right.left = other.min_node.left
|
||||
other.min_node.left.right = self.min_node.right
|
||||
self.min_node.right = other.min_node
|
||||
other.min_node.left = self.min_node
|
||||
|
||||
if other.min_node.val < self.min_node.val:
|
||||
self.min_node = other.min_node
|
||||
|
||||
self.size += other.size
|
||||
|
||||
def __link_trees(self, node1, node2):
|
||||
"""
|
||||
Links two trees of same degree.
|
||||
|
||||
Args:
|
||||
node1: First tree's root node.
|
||||
node2: Second tree's root node.
|
||||
"""
|
||||
node1.remove()
|
||||
if node2.child:
|
||||
node2.child.add_sibling(node1)
|
||||
else:
|
||||
node2.child = node1
|
||||
node1.parent = node2
|
||||
node2.degree += 1
|
||||
node1.mark = False
|
||||
|
||||
def delete_min(self):
|
||||
"""
|
||||
Removes and returns the minimum value from the heap.
|
||||
|
||||
Returns:
|
||||
The minimum value that was removed.
|
||||
|
||||
Raises:
|
||||
IndexError: If the heap is empty.
|
||||
"""
|
||||
if not self.min_node:
|
||||
raise IndexError("Heap is empty")
|
||||
|
||||
min_val = self.min_node.val
|
||||
|
||||
# Add all children to root list
|
||||
if self.min_node.child:
|
||||
curr = self.min_node.child
|
||||
while True:
|
||||
next_node = curr.right
|
||||
curr.parent = None
|
||||
curr.mark = False
|
||||
self.min_node.add_sibling(curr)
|
||||
if curr.right == self.min_node.child:
|
||||
break
|
||||
curr = next_node
|
||||
|
||||
# Remove minimum node
|
||||
if self.min_node.right == self.min_node:
|
||||
self.min_node = None
|
||||
else:
|
||||
self.min_node.remove()
|
||||
self.min_node = self.min_node.right
|
||||
self.__consolidate()
|
||||
|
||||
self.size -= 1
|
||||
return min_val
|
||||
|
||||
def __consolidate(self):
|
||||
"""
|
||||
Consolidates the trees in the heap after a delete_min operation.
|
||||
|
||||
This is an internal method that maintains the heap's structure.
|
||||
"""
|
||||
max_degree = int(self.size**0.5) + 1
|
||||
degree_table = [None] * max_degree
|
||||
|
||||
# Collect all roots
|
||||
roots = []
|
||||
if self.min_node:
|
||||
current_root = self.min_node
|
||||
curr = self.min_node
|
||||
while True:
|
||||
roots.append(current_root)
|
||||
if current_root.right == self.min_node:
|
||||
roots.append(curr)
|
||||
curr = curr.right
|
||||
if curr == self.min_node:
|
||||
break
|
||||
current_root = current_root.right
|
||||
|
||||
for current_root in roots:
|
||||
root_node = current_root
|
||||
current_degree = root_node.degree
|
||||
# Consolidate trees
|
||||
for root in roots:
|
||||
degree = root.degree
|
||||
while degree_table[degree]:
|
||||
other = degree_table[degree]
|
||||
if root.val > other.val:
|
||||
root, other = other, root
|
||||
self.__link_trees(other, root)
|
||||
degree_table[degree] = None
|
||||
degree += 1
|
||||
degree_table[degree] = root
|
||||
|
||||
while degree_table[current_degree] is not None:
|
||||
other_root = degree_table[current_degree]
|
||||
if root_node.key > other_root.key:
|
||||
root_node, other_root = other_root, root_node
|
||||
|
||||
other_root.left.right = other_root.right
|
||||
other_root.right.left = other_root.left
|
||||
|
||||
if root_node.child.key is None:
|
||||
root_node.child = other_root
|
||||
other_root.right = other_root
|
||||
other_root.left = other_root
|
||||
else:
|
||||
self._insert_into_circular_list(root_node.child, other_root)
|
||||
|
||||
other_root.parent = root_node
|
||||
root_node.degree += 1
|
||||
other_root.marked = False
|
||||
|
||||
degree_table[current_degree] = Node(None)
|
||||
current_degree += 1
|
||||
|
||||
degree_table[current_degree] = root_node
|
||||
|
||||
self.min_node.key = None
|
||||
# Find new minimum
|
||||
self.min_node = None
|
||||
for degree in range(max_degree):
|
||||
if degree_table[degree] is not None and (
|
||||
self.min_node is None or (degree_table[degree] < self.min_node.key)
|
||||
):
|
||||
if degree_table[degree]:
|
||||
if not self.min_node:
|
||||
self.min_node = degree_table[degree]
|
||||
self.min_node.left = self.min_node
|
||||
self.min_node.right = self.min_node
|
||||
else:
|
||||
self.min_node.add_sibling(degree_table[degree])
|
||||
if degree_table[degree].val < self.min_node.val:
|
||||
self.min_node = degree_table[degree]
|
||||
|
||||
def decrease_key(self, node: Node, new_key: float | None) -> None:
|
||||
"""Decrease the key value of a given node.
|
||||
|
||||
This operation updates the key of a node to a new, smaller value and
|
||||
maintains the min-heap property by potentially cutting the node from
|
||||
its parent and performing cascading cuts up the tree.
|
||||
def decrease_key(self, node, new_val):
|
||||
"""
|
||||
Decreases the value of a node.
|
||||
|
||||
Args:
|
||||
node: The node whose key should be decreased
|
||||
new_key: The new key value, must be smaller than the current key
|
||||
node: The node whose value should be decreased.
|
||||
new_val: The new value for the node.
|
||||
|
||||
Example:
|
||||
>>> heap = FibonacciHeap()
|
||||
>>> node1 = heap.insert(5)
|
||||
>>> heap.decrease_key(node, 3)
|
||||
>>> node.key
|
||||
3
|
||||
>>> heap.find_min()
|
||||
3
|
||||
>>> heap.decrease_key(node, 1)
|
||||
>>> node.key
|
||||
1
|
||||
>>> heap.find_min()
|
||||
1
|
||||
Raises:
|
||||
ValueError: If new value is greater than current value.
|
||||
"""
|
||||
if new_key > node.key:
|
||||
raise ValueError("New key is greater than current key")
|
||||
if new_val > node.val:
|
||||
raise ValueError("New value is greater than current value")
|
||||
|
||||
node.key = new_key
|
||||
parent_node = node.parent
|
||||
node.val = new_val
|
||||
parent = node.parent
|
||||
|
||||
if parent_node.key is not None and node.key < parent_node.key:
|
||||
self._cut(node, parent_node)
|
||||
self._cascading_cut(parent_node)
|
||||
if parent and node.val < parent.val:
|
||||
self.__cut(node, parent)
|
||||
self.__cascading_cut(parent)
|
||||
|
||||
if node.key < self.min_node.key:
|
||||
if node.val < self.min_node.val:
|
||||
self.min_node = node
|
||||
|
||||
def _cut(self, child_node: Node, parent_node: Node) -> None:
|
||||
"""Cut a node from its parent and add it to the root list.
|
||||
|
||||
This is a helper method used in decrease_key operations. When a node's key
|
||||
becomes smaller than its parent's key, it needs to be cut from its parent
|
||||
and added to the root list to maintain the min-heap property.
|
||||
def __cut(self, node, parent):
|
||||
"""
|
||||
Cuts a node from its parent.
|
||||
|
||||
Args:
|
||||
child_node: The node to be cut from its parent
|
||||
parent_node: The parent node from which to cut
|
||||
node: Node to be cut.
|
||||
parent: Parent of the node to be cut.
|
||||
""" """
|
||||
Performs cascading cut operation.
|
||||
|
||||
Note:
|
||||
This is an internal method that maintains heap properties during
|
||||
decrease_key operations. It should not be called directly from
|
||||
outside the class.
|
||||
Args:
|
||||
node: Starting node for cascading cut.
|
||||
"""
|
||||
if child_node.right == child_node:
|
||||
parent_node.child = Node(None)
|
||||
|
||||
parent.degree -= 1
|
||||
if parent.child == node:
|
||||
parent.child = node.right if node.right != node else None
|
||||
node.remove()
|
||||
node.left = node
|
||||
node.right = node
|
||||
node.parent = None
|
||||
node.mark = False
|
||||
self.min_node.add_sibling(node)
|
||||
|
||||
def __cascading_cut(self, node):
|
||||
"""
|
||||
Performs cascading cut operation.
|
||||
|
||||
Args:
|
||||
node: Starting node for cascading cut.
|
||||
"""
|
||||
|
||||
if parent := node.parent:
|
||||
if not node.mark:
|
||||
node.mark = True
|
||||
else:
|
||||
parent_node.child = child_node.right
|
||||
child_node.right.left = child_node.left
|
||||
child_node.left.right = child_node.right
|
||||
self.__cut(node, parent)
|
||||
self.__cascading_cut(parent)
|
||||
|
||||
parent_node.degree -= 1
|
||||
|
||||
self._insert_into_circular_list(self.min_node, child_node)
|
||||
child_node.parent = Node(None)
|
||||
child_node.marked = False
|
||||
|
||||
def _cascading_cut(self, current_node: Node) -> None:
|
||||
"""Perform cascading cut operation.
|
||||
|
||||
Args:
|
||||
current_node: The node to start cascading cut from
|
||||
def __str__(self):
|
||||
"""
|
||||
if (parent_node := current_node.parent) is not None:
|
||||
if not current_node.marked:
|
||||
current_node.marked = True
|
||||
else:
|
||||
self._cut(current_node, parent_node)
|
||||
self._cascading_cut(parent_node)
|
||||
|
||||
def delete(self, node: Node) -> None:
|
||||
"""Delete a node from the heap.
|
||||
|
||||
This operation removes a given node from the heap by first decreasing
|
||||
its key to negative infinity (making it the minimum) and then extracting
|
||||
the minimum.
|
||||
|
||||
Args:
|
||||
node: The node to be deleted from the heap
|
||||
|
||||
Example:
|
||||
>>> heap = FibonacciHeap()
|
||||
>>> node1 = heap.insert(3)
|
||||
>>> node2 = heap.insert(2)
|
||||
>>> heap.delete(node1)
|
||||
>>> heap.find_min()
|
||||
2
|
||||
>>> heap.total_nodes
|
||||
1
|
||||
|
||||
Note:
|
||||
This operation has an amortized time complexity of O(log n)
|
||||
as it combines decrease_key and extract_min operations.
|
||||
"""
|
||||
self.decrease_key(node, float("-inf"))
|
||||
self.extract_min()
|
||||
|
||||
def find_min(self) -> float | None:
|
||||
"""Return the minimum key without removing it from the heap.
|
||||
|
||||
This operation provides quick access to the minimum key in the heap
|
||||
without modifying the heap structure.
|
||||
Returns a string representation of the heap.
|
||||
|
||||
Returns:
|
||||
float | None: The minimum key value, or None if the heap is empty
|
||||
|
||||
Example:
|
||||
>>> heap = FibonacciHeap()
|
||||
>>> heap.find_min() is None
|
||||
True
|
||||
>>> node1 = heap.insert(3)
|
||||
>>> heap.find_min()
|
||||
3
|
||||
str: A string showing the heap structure.
|
||||
"""
|
||||
return self.min_node.key if self.min_node else Node(None).key
|
||||
if not self.min_node:
|
||||
return "Empty heap"
|
||||
|
||||
def is_empty(self) -> bool:
|
||||
"""Check if heap is empty.
|
||||
def print_tree(node, level=0):
|
||||
result = []
|
||||
curr = node
|
||||
while True:
|
||||
result.append("-" * level + str(curr.val))
|
||||
if curr.child:
|
||||
result.extend(print_tree(curr.child, level + 1))
|
||||
curr = curr.right
|
||||
if curr == node:
|
||||
break
|
||||
return result
|
||||
|
||||
Returns:
|
||||
bool: True if heap is empty, False otherwise
|
||||
|
||||
Examples:
|
||||
>>> heap = FibonacciHeap()
|
||||
>>> heap.is_empty()
|
||||
True
|
||||
>>> node = heap.insert(1)
|
||||
>>> heap.is_empty()
|
||||
False
|
||||
"""
|
||||
return self.min_node.key is None
|
||||
|
||||
def merge(self, other_heap: FibonacciHeap) -> None:
|
||||
"""Merge another Fibonacci heap into this one.
|
||||
|
||||
This operation combines two Fibonacci heaps by concatenating their
|
||||
root lists and updating the minimum pointer if necessary. The other
|
||||
heap is effectively consumed in this process.
|
||||
|
||||
Args:
|
||||
other_heap: Another FibonacciHeap instance to merge into this one
|
||||
|
||||
Example:
|
||||
>>> heap1 = FibonacciHeap()
|
||||
>>> node1 = heap1.insert(3)
|
||||
>>> heap2 = FibonacciHeap()
|
||||
>>> node2 = heap2.insert(2)
|
||||
>>> heap1.merge(heap2)
|
||||
>>> heap1.find_min()
|
||||
2
|
||||
>>> heap1.total_nodes
|
||||
2
|
||||
"""
|
||||
if other_heap.min_node.key is None:
|
||||
return
|
||||
if self.min_node.key is None:
|
||||
self.min_node = other_heap.min_node
|
||||
else:
|
||||
self.min_node.right.left = other_heap.min_node.left
|
||||
other_heap.min_node.left.right = self.min_node.right
|
||||
self.min_node.right = other_heap.min_node
|
||||
other_heap.min_node.left = self.min_node
|
||||
|
||||
if other_heap.min_node.key < self.min_node.key:
|
||||
self.min_node = other_heap.min_node
|
||||
|
||||
self.total_nodes += other_heap.total_nodes
|
||||
return "\n".join(print_tree(self.min_node))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
|
Loading…
Reference in New Issue
Block a user