Merge pull request #1 from jalving/master

Update to Plasmo v0.3.0

Project.toml

@@ -17,6 +17,7 @@ SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 
 [compat]
 julia = "1"
+Plasmo = "0.3.0"
 
 [extras]
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"

README.md

@@ -1,26 +1,33 @@
 # PipsSolver.jl
 
-[![Build Status](https://travis-ci.com/jalving/MGPipsSolver.jl.svg?branch=master)](https://travis-ci.com/jalving/PipsSolver.jl)
-[![Codecov](https://codecov.io/gh/jalving/MGPipsSolver.jl/branch/master/graph/badge.svg)](https://codecov.io/gh/jalving/PipsSolver.jl)
 
 ## Overview
 PipsSolver.jl is a Julia interface to the [PIPS-NLP](https://github.com/Argonne-National-Laboratory/PIPS/tree/master/PIPS-NLP) nonlinear optimization solver.
 Running the solver requires a working PIPS-NLP installation following the [instructions](https://github.com/Argonne-National-Laboratory/PIPS).  
 The PipsSolver.jl package works with the graph-based algebraic modeling package [Plasmo.jl](https://github.com/zavalab/Plasmo.jl).
 
 ## Installation
-PipsSolver.jl can be installed using the following Julia Pkg command.
+PipsSolver.jl can be installed using the following Julia Pkg command. Note that currently, PipsSolver.jl only works with Plasmo.jl v0.3.0.
 
 ```julia
 using Pkg
+Pkg.add(Pkg.PackageSpec(name="Plasmo", version="0.3.0"))
 Pkg.add(PackageSpec(url="https://github.com/zavalab/PipsSolver.jl.git"))
 ```
 
 ## Useage
 ```julia
 using MPIClusterManagers # to import MPIManager
 using Distributed   # need to also import Distributed to use addprocs()
-using Plasmo
+
+#Setup worker environments
+#This will load the environment specified in this script's directory onto each worker
+@everywhere using Pkg
+@everywhere Pkg.activate((@__DIR__))
+
+#Load Plasmo and PipsSolver on every worker
+@everywhere using Plasmo
+@everywhere using PipsSolver
 
 graph = OptiGraph()
 
@@ -47,21 +54,12 @@ manager=MPIManager(np=2) # specify, number of mpi workers, launch cmd, etc.
 addprocs(manager)        # start mpi workers and add them as julia workers too.
 
 
-#Setup worker environments
-#NOTE: You will need to load your Julia environment onto each worker
-@everywhere using Pkg
-@everywhere Pkg.activate((@__DIR__))
-
-#Load Plasmo and PipsSolver on every worker
-@everywhere using Plasmo
-@everywhere using PipsSolver
-
-
-julia_workers = sort(collect(values(manager.mpi2j))) # #Distribute the graph to workers
-
-remote_references = PipsSolver.distribute(graph,julia_workers,remote_name = :pipsgraph)  #create the variable pipsgraph on each worker
+#Map julia workers to MPI ranks
+julia_workers = sort(collect(values(manager.mpi2j)))
+#Distribute the graph to workers. This creates the variable `pipsgraph` on each worker with an allocation of optinodes.
+remote_references = PipsSolver.distribute(graph,julia_workers,remote_name = :pipsgraph)
 
-# The remote modelgraphs can be queried if they are fetched from the other workers
+# The remote optigraphs can be queried if they are fetched from the other workers.  
 r1 = fetch(remote_references[1])
 r2 = fetch(remote_references[2])
 

src/PipsNlpInterface.jl

@@ -70,11 +70,11 @@ function getData(m::JuMP.Model)
     end
 end
 
-function pipsnlp_solve(graph::ModelGraph) #Assume graph variables and constraints are first stage
+function pipsnlp_solve(graph::OptiGraph) #Assume graph variables and constraints are first stage
 
     #TODO SUBGRAPHS with linkconstraints to subnodes
     if has_subgraphs(graph)
-        error("The PIPS-NLP does not yet support ModelGraphs with subgraphs.  You will need to aggregate the graph before calling pipsnlp_solve")
+        error("The PIPS-NLP does not yet support OptiGraphs with subgraphs.  You will need to aggregate the graph before calling pipsnlp_solve")
     end
 
     if has_NLlinkconstraints(graph)
@@ -711,7 +711,7 @@ function pipsnlp_solve(graph::ModelGraph) #Assume graph variables and constraint
         println("PIPS-NLP time:   ",  time() - t1, " (s)")
     end
 
-    #TODO.  Put solution onto actual ModelNode
+    #TODO.  Put solution onto actual OptiNode
     for (idx,node) in enumerate(modelList)  #set solution values for each model
         local_data = getData(node)
         if idx != 1

src/distribute.jl

@@ -1,5 +1,5 @@
 #Distribute a modelgraph among workers.  Each worker should have the same master model.  Each worker will be allocated some of the nodes in the original modelgraph
-function distribute(mg::ModelGraph,to_workers::Vector{Int64};remote_name = :graph)
+function distribute(mg::OptiGraph,to_workers::Vector{Int64};remote_name = :graph)
     #NOTE: Linkconstraints keep their indices in new graphs, NOTE: Link constraint row index needs to match on each worker
     #NOTE: Does not yet support subgraphs.  Aggregate first
     #Create remote channel to store the nodes we want to send
@@ -9,7 +9,7 @@ function distribute(mg::ModelGraph,to_workers::Vector{Int64};remote_name = :grap
     channel_indices = RemoteChannel(1)
 
     to_workers = sort(to_workers)
-    
+
     n_nodes = getnumnodes(mg)
     n_workers = length(to_workers)
     nodes_per_worker = Int64(floor(n_nodes/n_workers))
@@ -84,10 +84,10 @@ function distribute(mg::ModelGraph,to_workers::Vector{Int64};remote_name = :grap
     return remote_references
 end
 
-function _create_worker_modelgraph(modelnodes::Vector{ModelNode},node_indices::Vector{Int64},n_nodes::Int64,n_linkeq_cons::Int64,n_linkineq_cons::Int64,
+function _create_worker_modelgraph(modelnodes::Vector{OptiNode},node_indices::Vector{Int64},n_nodes::Int64,n_linkeq_cons::Int64,n_linkineq_cons::Int64,
     link_ineq_lower::Vector,link_ineq_upper::Vector)
-    graph = ModelGraph()
-    graph.node_idx_map = Dict{ModelNode,Int64}()
+    graph = OptiGraph()
+    graph.node_idx_map = Dict{OptiNode,Int64}()
 
     #Add nodes to worker's graph.  Each worker should have the same number of nodes, but some will be empty.
     for i = 1:n_nodes
@@ -153,7 +153,7 @@ function _create_worker_modelgraph(modelnodes::Vector{ModelNode},node_indices::V
     return graph
 end
 
-function _add_linkeq_terms(modelnodes::Vector{ModelNode})
+function _add_linkeq_terms(modelnodes::Vector{OptiNode})
     linkeqconstraints = OrderedDict()
     for node in modelnodes
         partial_links = node.partial_linkeqconstraints
@@ -174,7 +174,7 @@ function _add_linkeq_terms(modelnodes::Vector{ModelNode})
     return linkeqconstraints
 end
 
-function _add_linkineq_terms(modelnodes::Vector{ModelNode})
+function _add_linkineq_terms(modelnodes::Vector{OptiNode})
     linkineqconstraints = OrderedDict()
     for node in modelnodes
         partial_links = node.partial_linkineqconstraints