Hungarian

The package provides one implementation of the Hungarian algorithm(Kuhn-Munkres algorithm) based on its matrix interpretation. This implementation uses a sparse matrix to keep tracking those marked zeros, so it costs less time and memory than Munkres.jl. Benchmark details can be found here.

Installation

pkg> add Hungarian

Quick start

Let's say we have 5 workers and 3 tasks with the following cost matrix:

weights = [ 24     1     8;
             5     7    14;
             6    13    20;
            12    19    21;
            18    25     2]

We can solve the assignment problem by:

julia> using Hungarian

julia> assignment, cost = hungarian(weights)
([2, 1, 0, 0, 3], 8)

# worker 1 => task 2 with weights[1,2] = 1
# worker 2 => task 1 with weights[2,1] = 5
# worker 5 => task 3 with weights[5,3] = 2
# the minimal cost is 1 + 5 + 2 = 8  

Since each worker can perform only one task and each task can be assigned to only one worker, those 0s in the assignment mean that no task is assigned to those workers.

If a job-worker assignment is not possible, use the special missing value to indicate which pairs are disallowed:

julia> using Hungarian

julia> weights = [missing 1 1; 1 0 1; 1 1 0]
3×3 Matrix{Union{Missing, Int64}}:
  missing  1  1
 1         0  1
 1         1  0

julia> assignment, cost = hungarian(weights)
([2, 1, 3], 2)

NOTE: This package does not support Inf weights. All Infs are converted to prevfloat(typemax(T)) before running the munkres algorithm.

Usage

When solving a canonical assignment problem, namely, the cost matrix is square, one can directly get the matching via Hungarian.munkres(x) instead of hungarian(x):

julia> using Hungarian

julia> matching = Hungarian.munkres(rand(5,5))
5×5 SparseArrays.SparseMatrixCSC{Int8, Int64} with 9 stored entries:
 1  2  ⋅  ⋅  ⋅
 ⋅  ⋅  1  ⋅  2
 2  ⋅  ⋅  ⋅  ⋅
 1  ⋅  2  ⋅  ⋅
 ⋅  ⋅  ⋅  2  1

# 0 => non-zero
# 1 => zero
# 2 => STAR
julia> Matrix(matching)
5×5 Matrix{Int8}:
 1  2  0  0  0
 0  0  1  0  2
 2  0  0  0  0
 1  0  2  0  0
 0  0  0  2  1

julia> [findfirst(matching[i,:].==Hungarian.STAR) for i = 1:5]
5-element Vector{Int64}:
 2
 5
 1
 3
 4

julia> [findfirst(matching[:,i].==Hungarian.STAR) for i = 1:5]
5-element Vector{Int64}:
 3
 1
 4
 5
 2

If a job-worker assignment is not possible, use the special missing value to indicate which pairs are disallowed:

julia> using Hungarian

julia> weights = [missing 1 1; 1 0 1; 1 1 0]
3×3 Matrix{Union{Missing, Int64}}:
  missing  1  1
 1         0  1
 1         1  0

julia> matching = Hungarian.munkres(weights)
3×3 SparseArrays.SparseMatrixCSC{Int8, Int64} with 6 stored entries:
 ⋅  2  1
 2  1  ⋅
 1  ⋅  2

References

1. J. Munkres, "Algorithms for the Assignment and Transportation Problems", Journal of the Society for Industrial and Applied Mathematics, 5(1):32–38, 1957 March.

2. http://csclab.murraystate.edu/bob.pilgrim/445/munkres.html