Skip to content

soniakeys/fib

Repository files navigation

Fib

Package Fib implements a Fibonacci heap.

It mostly follows Fredman and Tarjan’s "Fibonacci Heaps and Their Uses in Improved Network Optimization Algorithms", Journal of the Association for Computing Machinery, Vol. 34, No. 3, July 1987.

It is public domain and has 100% test coverage.

Some implementation notes

A significant enhancement is that while Fredman and Tarjan describe storing "items, each with a real-valued key," this implementation stores any orderable type implementing a less-than method. It does not depend on floating point comparisons.

This implementation does not maintain a count of the number of values present in the heap. F&T describe one use for the count, for sizing a certain array by log(count), but this implementation uses a Go map for reasons of simplicity and robustness and does not need the count.

Popular Fibonacci heap implementations follow Cormen, Leiserson, Rivest, and Stein (CLRS) in "Introduction to Algorithms." This package follows the 1987 paper by Fredman and Tarjan. A significant difference is in their algorithms for the delete function. Fredman and Tarjan’s is "lazier."

Compared to the standard libarary container/heap

Use the standard library!

The standard library heap is a few times better in both speed and space. That’s typical between Fibonacci heaps and array-based binary heaps. If you read that Fibonacci heaps have better asymptotic performance and you think you have a big data set and so you n e e d a Fibonacci heap, you are probably wrong. In practice Fibonacci heaps rarely outperform simpler heaps. I was actually really pleased that this implementation came out only a few times slower than the standard library.

If you don’t care and you want one anyway, this package may serve you well. After all, a few times really fast is still really fast. Still, there are some differences in functionality and in the API.

For container/heap you must implement the five methods of heap.Interface. For this fib package, you also implement an interface, but there’s just a single method to implement.

Table 1. Table Functions and methods
operation container/heap fib

Heap construction

h := some indexable container that satisfies heap.Interface.

h := &Heap{} or h := new(fib.Heap)

"Heapify" populated container

heap.Init(h)

N/A

Put a new value on the heap

heap.Push(x)

h.Insert(x)

Test if heap is empty

h.Len() == 0

h.Node == nil

Number of values on heap

h.Len()

N/A

"Top" of heap, minimum value, next in priorty

h[0] (if you used a slice)

h.Min() or h.Node.Value()

Take min value from heap

heap.Pop(h)

h.DeleteMin()

Decrease heaped value

heap.Fix(h, i)

h.DecreaseKey(n)

Change heaped value

heap.Fix(h, i)

N/A

Remove a value from heap

heap.Remove(h, i)

h.Remove(n)

Merge two heaps

N/A

h.Meld(h2)