Version Problems?

[howard4086]

I have been using JuMP to solve a MILP (with CBC), and it seems the latest version doesn't work with my code. When I tried to create a model with Cbc as the solver the code below returned an error:
m=Model(solver=CbcSolver(MipMaxSeconds=30,MipFractionalGap=MIPGAP,LogLevel=CBCLogLevel))

This would not work with the latest version of Clp/Cbc, so I pinned it back and it works fine now. Also - I'm use Julia Studio, so I'm on v0.2.0 of Julia (perhaps that is the problem?) I didn't know if the command has changed in the latest version of Clp or JuMP, or if it is a compatibility error with my older version of Julia? Its not a huge deal as it works with the older packages, but I wanted to understand why it was necessary.

Here are my Package versions: (that work with the command above)

• Clp 0.0.8 pinned.aac279f6.tmp
• JuMP 0.5.5
• BinDeps 0.3.5
• Calculus 0.1.3
• Cbc 0.0.7

Thanks!

-Nick

[mlubin]

What's the error that you're receiving? The names of the Cbc parameters have recently changed, take a look at: https://github.com/JuliaOpt/Cbc.jl. The latest version of JuMP/Clp/Cbc requires julia 0.3.0. If you're on 0.2, you should still have a working set of packages though.

[howard4086]

When I use the latest packages I get the following:
LoadError("MIP_Generic.jl",106,ErrorException("CbcSolver not defined"))

[howard4086]

Perhaps an incompatibility with Julia 0.2.0?

[mlubin]

Is there any error when you say using Cbc in a clean Julia session?

[howard4086]

No there isn't, but I gave the wrong error (when I freed the package I hadn't re-built Cbc). using Cbc works fine, and m=Model() works fine, but trying to specify the solver in any way to be CbcSolver results in:
LoadError("MIP_Generic.jl",106,MethodError(Model,(0,:Min,[],[],[],0,[],[],[],[],0,[],[],[],nothing,CbcSolver({(:MipMaxSeconds,30.0),(:MipFractionalGap,0.01),(:LogLevel,0)}),false,nothing,nothing,nothing,[],IndexedVector{Float64}([],[],0),nothing,Dict{Symbol,Any}())))

[mlubin]

I wasn't able to reproduce this issue on a clean install of julia 0.2. You might have a weird mix of package versions installed. Could you try clearing your packages (delete the directory reported by Pkg.dir()) and installing a fresh copy of everything? You shouldn't need to pin any versions.

[howard4086]

I cleared it and reinstalled all the packages and still get the error if I don't pin Clp. I wonder if its due to the julia-readline.exe interpreter that Julia Studio uses?

[joehuchette]

I think that JuliaStudio isn't handling the errors particularly gracefully, which is why you see the LoadError. There's definitely an error propagating up from the library, though.

[joehuchette]

@howard4086 could you report what version of MathProgBase you have installed? I'm wondering if you're running a 0.3.0-dev version.

[howard4086]

Here is my current set of packages:

• Clp 0.0.8 pinned.aac279f6.tmp
• JuMP 0.5.5
• Logging 0.0.4
• PyCall 0.4.9
• BinDeps 0.3.5
• Calculus 0.1.3
• Cbc 0.0.7
• DataStructures 0.2.15
• DualNumbers 0.1.0
• Graphs 0.3.1
• MathProgBase 0.2.5
• ReverseDiffSparse 0.1.2
• SHA 0.0.3
• URIParser 0.0.2

[joehuchette]

I think what's going on is that there's an incompatibility with the MathProgBase v0.2.5 interface and the way Clp v0.0.8 tries to plug into it (this manifests itself with the no-method error you reported above). METADATA should be changed to reflect this, which would mean Clp v0.0.7 is an upper bound for your set up and which (I hope!) should fix this.

[mlubin]

On a clean 0.2:

julia> Pkg.status()
Required packages:
 - Cbc                           0.0.8
 - Clp                           0.0.9
 - JuMP                          0.5.7
Additional packages:
 - BinDeps                       0.3.5
 - Calculus                      0.1.3
 - DataStructures                0.2.15
 - DualNumbers                   0.1.0
 - Graphs                        0.3.1
 - MathProgBase                  0.3.0
 - ReverseDiffSparse             0.1.2
 - SHA                           0.0.3
 - URIParser                     0.0.2

The solver versions are compatible with MPB 0.3.0. @howard4086, could you post Pkg.status() with everything unpinned? Also a small piece of code that reproduces the issue would be helpful.

[howard4086]

I'm using Julia 0.2, but through Julia Studio, which actually calls julia-readline.exe.

Here is my Pkg.status():
Required packages:

• Clp 0.0.9
• JuMP 0.5.7
• Logging 0.0.4
• PyCall 0.4.9
  Additional packages:
• BinDeps 0.3.5
• Calculus 0.1.3
• Cbc 0.0.8
• DataStructures 0.2.15
• DualNumbers 0.1.0
• Graphs 0.3.1
• MathProgBase 0.3.0
• ReverseDiffSparse 0.1.2
• SHA 0.0.3
• URIParser 0.0.2

Then here is the code:
using Cbc
using JuMP
m=Model(solver=CbcSolver())

This generates:
MethodError(Model,(0,:Min,[],[],[],0,[],[],[],[],0,[],[],[],nothing,CbcSolver({}),false,nothing,nothing,nothing,[],IndexedVector{Float64}([],[],0),nothing,Dict{Symbol,Any}()))

[mlubin]

I think I've fixed the issue, try Pkg.checkout("JuMP", "release-0.5"). We can release JuMP 0.5.8 for this.

[howard4086]

When I run:
Pkg.checkout("JuMP", "release-0.5")

I get:
INFO: Checking out JuMP release-0.5...
error: pathspec 'release-0.5' did not match any file(s) known to git.
ErrorException("failed process: Process(git '--work-tree=\\.julia\\JuMP' '--git-dir=\\.julia\\JuMP\\.git' checkout -q release-0.5, ProcessExited(1)) [1]")

[mlubin]

Try Pkg.update(); Pkg.checkout("JuMP, "b4c79f4ce6")

[howard4086]

I get this error:

INFO: Updating METADATA...
INFO: Computing changes...
INFO: No packages to install, update or remove.
INFO: Checking out JuMP b4c79f4...
error: pathspec 'b4c79f4ce6' did not match any file(s) known to git.

[mlubin]

Not sure why that's not working, but I just released JuMP 0.5.8 with the fix. Try Pkg.update() now.

[howard4086]

Good news is that it now works! Thanks!

However, it appears the latest version has changed the syntax of Variable?

I have been using Variable() to create my variables one at a time as I am using dictionaries with indices not static at compile time based on the problem data files (which can change depending on the data loaded). In the past this worked just fine using:
Variable(m,0,10,1) or Variable(m,0,Inf,0) for an integer or continuous variable for example.

Now when I call this I get:
MethodError(Variable,(Feasibility problem with:

• 0 linear constraints
• 0 variables
  Solver set to Cbc,0,Inf,0)))

What am I doing wrong here?

Thanks!

[mlubin]

@howard4086 The final argument for Variable is now :Int, :Bin, or :Cont for integer, binary, and continuous variables respectively. We forgot to update the documentation: #272.

But anyway there's no restriction on variable indices being defined at compile time. You can do something like:

S = {}
# fill S using input data
@defVar(m, x[S])

It shouldn't be necessary to use Variable().

[howard4086]

Thats very nice - thanks. After fixing my variable declarations everything worked, except I found that CbcSolver did not respond to any options I specified:

m=Model(solver=CbcSolver(MipMaxSeconds=30,MipFractionalGap=.01,LogLevel=1)

This command ran and solved fine, but it solved to optimality, with the default LogLevel and took 72 seconds.

I found that the CoinMP.dll file was dated 9/13/2013, so I rolled it forward to a version dated 1/7/2014, and then everything worked fine. This morning I had cleared all of my packages and reinstalled clean, so the 9/13/2013 version is the default one it pulled down - perhaps this is a mistake and you should have the newer version instead?

Either way everything works well now with the latest packages. Thanks so much for the help!

[mlubin]

The latest version of the Cbc wrapper (which requires julia 0.3) has newly updated binaries and renamed parameters, see https://github.com/JuliaOpt/Cbc.jl. This is just as a heads up for when you eventually update to julia 0.3.