Ensure that random_kruskal_mst doesn't generate the same spanning tre…

…e every time (#120)

* Ensure that random_kruskal_mst doesn't generate the same spanning tree every time

* Stop running tests for macOS

As it appears that this is an upstream issue with installation on macOS,
it's probably best to remove the macOS test for now until it gets
resolved. In the meantime, tests for Windows and Linux should suffice.

* Add documentation about the AbstractRNG type and update type annotations

* Add example of AbstractRNG type to recom func

.github/workflows/test.yml

@@ -7,8 +7,8 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        julia-version: ['1.5.3']
-        os: [ubuntu-latest, windows-latest, macos-10.15]
+        julia-version: ['1.6.1']
+        os: [ubuntu-latest, windows-latest]
 
     steps:
     - uses: actions/checkout@v2

src/balance_edges.jl

@@ -52,7 +52,7 @@ end
     random_kruskal_mst(graph::BaseGraph,
                        edges::Array{Int, 1},
                        nodes::Array{Int, 1},
-                       rng=MersenneTwister(1234))::BitSet
+                       rng::AbstractRNG=Random.default_rng())
 
 Generates and returns a random minimum spanning tree from the subgraph induced
 by `edges` and `nodes`, using Kruskal's MST algorithm.
@@ -65,14 +65,15 @@ The `graph` represents the entire graph of the plan, where as `edges` and
 - graph: Underlying Graph object
 - edges: Array of edges of the sub-graph
 - nodes: Set of nodes of the sub-graph
+- rng: A random number generator that implements the [AbstractRNG type](https://docs.julialang.org/en/v1/stdlib/Random/#Random.AbstractRNG) (e.g. `Random.default_rng()` or `MersenneTwister(1234)`)
 
 *Returns* a BitSet of edges that form a mst.
 """
 function random_kruskal_mst(
     graph::BaseGraph,
     edges::Array{Int,1},
     nodes::Array{Int,1},
-    rng = MersenneTwister(1234),
+    rng::AbstractRNG = Random.default_rng(),
 )::BitSet
     weights = rand(rng, length(edges))
     return kruskal_mst(graph, edges, nodes, weights)

src/recom.jl

@@ -177,6 +177,9 @@ end
     - pop_constraint: PopulationConstraint to adhere to
     - num_tries:      num times to try getting a balanced cut from a subgraph
                       before giving up
+    - rng:            A random number generator that implements the 
+                      [AbstractRNG type](https://docs.julialang.org/en/v1/stdlib/Random/#Random.AbstractRNG) 
+                      (e.g. `Random.default_rng()` or `MersenneTwister(1234)`)
 """
 function get_valid_proposal(
     graph::BaseGraph,
@@ -189,7 +192,7 @@ function get_valid_proposal(
         D₁, D₂, sg_edges, sg_nodes = sample_subgraph(graph, partition, rng)
 
         for _ = 1:num_tries
-            mst_edges = random_kruskal_mst(graph, sg_edges, collect(sg_nodes))
+            mst_edges = random_kruskal_mst(graph, sg_edges, collect(sg_nodes), rng)
 
             # see if we can get a population-balanced cut in this mst
             proposal = get_balanced_proposal(
@@ -272,7 +275,9 @@ step of the chain.
                     representing the likelihood of accepting the
                     proposal. Should accept a `Partition` as input.
 - rng:              Random number generator. The user can pass in their
-                    own; otherwise, we use the default RNG from Random.
+                    own; otherwise, we use the default RNG from Random. Must
+                    implement the [AbstractRNG type](https://docs.julialang.org/en/v1/stdlib/Random/#Random.AbstractRNG) 
+                    (e.g. `Random.default_rng()` or `MersenneTwister(1234)`).
 - no\\_self\\_loops: If this is true, then a failure to accept a new state
                     is not considered a self-loop; rather, the chain
                     simply generates new proposals until the acceptance