matching algorithm for bipartite graphs (linear programming) #207

CarloLucibello · 2015-10-31T02:45:14Z

partially fixes #194

This is meant to be a temporary solution, exploiting the fact that for bipartite graphs matching can be relaxed to linear programming, untill someone finds the courage and the will to implement a more effcient matching algorithm.

For the time being I added JuMP to REQUIRE, I don't know how to set them as optional dependences.

CarloLucibello · 2015-10-31T03:21:27Z

looks like there is a problem on Julia 0.5

jpfairbanks · 2015-10-31T15:28:49Z

src/matching/linear-programming.jl

+end
+
+"""
+As `maximum_weigth_maximal_matching`, with the difference that the edges `e` with


weight not weigth

sbromberger · 2015-10-31T16:07:29Z

I don't like the idea of adding complex dependencies like JuMP for a single (set of) functions. If we can't eliminate the dependency, it looks like it might be better as a separate package.

jpfairbanks · 2015-11-01T03:18:08Z

@sbromberger, there are lots of graph problems that can be solved with a general purpose optimization tool like jump. Once we have easy access to such a tool it will be easier to add new functionality that leverages it.

Can we make it a Conditional dependency?

It plays a similar role to spectral methods, once you have a good linear solver on Graphs there are lots of ways to use it for analyzing graphs.

sbromberger · 2015-11-01T03:24:09Z

Conditional dependency is fine, but we should make sure that we've got more than one function that uses it.

CarloLucibello · 2015-11-01T03:42:06Z

Why we need more than one function? Having an algorithm for matching should be reason enough. IHaving an algorithm for matching is a worthwhile addition, it is an important problem in combinatorial optimization, it has a lot of applications, and on top of that few libreries iimplemnt it, so it adds value to LightGraphs

sbromberger · 2015-11-01T03:45:16Z

It just seems to me that requiring JuMP - which itself requires 4 different packages that are not part of LightGraphs REQUIRE (which themselves require 4+ packages that aren't part of LG REQUIRE) - just for a single function seems like it's too much extra baggage.

CarloLucibello · 2015-11-01T03:48:02Z

But conditional dependence means it should not go into REQUIRE, right?

sbromberger · 2015-11-01T03:48:39Z

That's right, and why I think we're ok with conditional dependence. But I think we still need to ensure we've got more than one function that will use it. This is a lot of baggage for a single enhancement.

CarloLucibello · 2015-11-01T03:50:11Z

So, should I modify the PR to make it conditional? and how?

sbromberger · 2015-11-01T03:51:03Z

I'd like @jpfairbanks to weigh in here – will this be useful for other functionality or is a JuMP conditional just good for matching?

sbromberger · 2015-11-01T03:53:24Z

(as for "how" - check out how we do it for GraphMatrices:

https://github.com/JuliaGraphs/LightGraphs.jl/blob/master/src/LightGraphs.jl#L12-L17 and
https://github.com/JuliaGraphs/LightGraphs.jl/blob/master/src/spectral.jl#L131-L135
)

codecov-io · 2015-11-01T03:55:12Z

Current coverage is `100.00%`

Merging #207 into master will not affect coverage as of 0249853

Powered by Codecov. Updated on successful CI builds.

CarloLucibello · 2015-11-03T12:25:01Z

If I don't add JuMP to REQUIRE, how do I say to Travis to add it for testing?

jpfairbanks · 2015-11-03T20:08:14Z

I think that matching is important and that using jump will support more functions in the future. You can have travis Pkg.add jump and the call the tests just like we do for graph matrices.

Once you have a powerful tool like jump, you find ways to use it. :)

sbromberger · 2015-11-03T20:38:45Z

src/matching/linear-programming.jl

+    status != :Optimal && error("Failure")
+    sol = getValue(x)
+
+    all(Bool[s == 1 || s == 0 for s in sol])


What is this statement supposed to do?

Since this is a continuous relaxation of an integer problem, here I check that the solution takes only integer values

But the statement doesn't do anything. It evaluates to true or false and then discards the answer.

oh, sure, that was meant to be in an assert. I'll fix it

CarloLucibello · 2015-11-04T14:50:52Z

this is ready for merge

sbromberger · 2015-11-04T20:20:38Z

src/LightGraphs.jl

+
+#matching
+maximum_weight_maximal_matching,
+


Is this a duplicate?

CarloLucibello · 2015-11-05T14:24:39Z

I renamed pi and \pi to m. I didn't remove the Int[] constructors, since I like them, they are much more concise then Vector{Int}(), and moreover they are used all over the place, e.g.

"""Add a new vertex to the graph `g`."""
function add_vertex!(g::SimpleGraph)
    n = length(vertices(g)) + 1
    g.vertices = 1:n
    push!(g.badjlist, Int[])
    push!(g.fadjlist, Int[])

    return n
end

sbromberger · 2015-11-05T16:30:09Z

I've been weeding them out; please file issues if you see Int[] constructors. See also https://groups.google.com/d/msg/julia-users/BPPjnk20JgY/Ihq6KMqkS-UJ.

sbromberger · 2015-11-05T21:42:50Z

@jpfairbanks - last call before merge. All good?

jpfairbanks · 2015-11-06T02:54:51Z

Go for it.
On Nov 5, 2015 4:42 PM, "Seth Bromberger" notifications@github.com wrote:

@jpfairbanks https://github.com/jpfairbanks - last call before merge.
All good?

—
Reply to this email directly or view it on GitHub
#207 (comment)
.

matching algorithm for bipartite graphs (linear programming)

sbromberger · 2015-11-06T17:00:26Z

All done. Thanks, @CarloLucibello and @jpfairbanks !!

sbromberger · 2015-11-11T23:25:21Z

I really think we need to revert this merge. It's causing travis to take > 10 mins to build all the JuMP dependencies, and we're seeing errors due to the macros and conditional dependencies.

I'm in favor of revisiting once we understand the impact, but requiring JuMP in LightGraphs is making it too top-heavy right now.

CarloLucibello · 2015-11-11T23:31:42Z

I'd say let's exclude matching just from testing, so that we don't also have to load JuMP. If this cannot be done easily, or if it would be bad practice to leave out from testing, let's just revert this. Also I don't know if it's possible to have codecov ignore the matching part, I wouldn't like it always complaining

jpfairbanks · 2015-11-12T01:28:34Z

Can we find a way to make travis faster with JuMP? What about storing all the dependencies prebuilt for travis? The JuMP repo uses this travis.yml file which installs glpk from the apt repo on linux.
This technique should reduce the travis build time.

language: julia
os:
    - linux
    - osx
julia:
    - 0.4
    - nightly
notifications:
    email: false
sudo: false
addons:
    apt_packages:
        - gfortran
        - liblapack-dev
        - libgmp-dev
        - libglpk-dev
script:
    - if [[ -a .git/shallow ]]; then git fetch --unshallow; fi
    - julia -e 'Pkg.clone(pwd())'
    - julia -e 'Pkg.test("JuMP", coverage=true)'
    - julia test/hygiene.jl
    - julia -e 'Pkg.add("Ipopt")' # needed for below tests
    - julia test/hockschittkowski/runhs.jl
after_success:
    - echo $TRAVIS_JULIA_VERSION
    - julia -e 'Pkg.add("Coverage"); cd(Pkg.dir("JuMP")); using Coverage; Coveralls.submit(process_folder()); Codecov.submit(process_folder())'

For more info: http://docs.travis-ci.com/user/installing-dependencies/

sbromberger · 2015-11-12T13:09:31Z

See https://groups.google.com/d/msg/julia-users/L1_E5uuSj0M/rdtJ_3mdBQAJ for a possible solution to the dependency issue.

jpfairbanks · 2015-11-13T00:01:22Z

yup that looks to be the right thing to allow LightGraphs to run without a dependence on JuMP. We should also use apt-get to test with JuMP on linux.

CarloLucibello added 9 commits October 30, 2015 15:40

add core_periphery_deg

5b1d083

add clustering coefficients

df5792c

implement bipartite_matching through linear programming

c61c757

add JuMP to require

5614d86

working on matching

deeef13

add funciton bipartite_map

a5fb287

finish work on matching

329aed2

add docs

1e0956f

Clp to REQUIRE

3cbd38f

jpfairbanks reviewed Oct 31, 2015
View reviewed changes

fix spelling mistake of weight

aced708

update test and other fixes

907800a

sbromberger reviewed Nov 3, 2015
View reviewed changes

CarloLucibello added 7 commits November 4, 2015 09:22

Clp to REQUIRE

257434b

fix spelling mistake of weight

3b52a21

update test and other fixes

2fdbdbb

optional dependence fot JuMP

8cc85ab

various fix following comment to PR

461df07

more descriptive error

ad916df

rebase

1b8219c

sbromberger reviewed Nov 4, 2015
View reviewed changes

some renmaings

751e72f

remove Int[] constructor

24122c0

sbromberger added a commit that referenced this pull request Nov 6, 2015

Merge pull request #207 from CarloLucibello/matching2

492523f

matching algorithm for bipartite graphs (linear programming)

sbromberger merged commit 492523f into sbromberger:master Nov 6, 2015

CarloLucibello deleted the matching2 branch November 7, 2015 15:48

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

matching algorithm for bipartite graphs (linear programming) #207

matching algorithm for bipartite graphs (linear programming) #207

CarloLucibello commented Oct 31, 2015

CarloLucibello commented Oct 31, 2015

jpfairbanks Oct 31, 2015

sbromberger commented Oct 31, 2015

jpfairbanks commented Nov 1, 2015

sbromberger commented Nov 1, 2015

CarloLucibello commented Nov 1, 2015

sbromberger commented Nov 1, 2015

CarloLucibello commented Nov 1, 2015

sbromberger commented Nov 1, 2015

CarloLucibello commented Nov 1, 2015

sbromberger commented Nov 1, 2015

sbromberger commented Nov 1, 2015

codecov-io commented Nov 1, 2015

CarloLucibello commented Nov 3, 2015

jpfairbanks commented Nov 3, 2015

sbromberger Nov 3, 2015

CarloLucibello Nov 4, 2015

sbromberger Nov 4, 2015

CarloLucibello Nov 4, 2015

CarloLucibello commented Nov 4, 2015

sbromberger Nov 4, 2015

CarloLucibello Nov 5, 2015

CarloLucibello commented Nov 5, 2015

sbromberger commented Nov 5, 2015

sbromberger commented Nov 5, 2015

jpfairbanks commented Nov 6, 2015

sbromberger commented Nov 6, 2015

sbromberger commented Nov 11, 2015

CarloLucibello commented Nov 11, 2015

jpfairbanks commented Nov 12, 2015

sbromberger commented Nov 12, 2015

jpfairbanks commented Nov 13, 2015


		#matching
		maximum_weight_maximal_matching,

matching algorithm for bipartite graphs (linear programming) #207

matching algorithm for bipartite graphs (linear programming) #207

Conversation

CarloLucibello commented Oct 31, 2015

CarloLucibello commented Oct 31, 2015

jpfairbanks Oct 31, 2015

Choose a reason for hiding this comment

sbromberger commented Oct 31, 2015

jpfairbanks commented Nov 1, 2015

sbromberger commented Nov 1, 2015

CarloLucibello commented Nov 1, 2015

sbromberger commented Nov 1, 2015

CarloLucibello commented Nov 1, 2015

sbromberger commented Nov 1, 2015

CarloLucibello commented Nov 1, 2015

sbromberger commented Nov 1, 2015

sbromberger commented Nov 1, 2015

codecov-io commented Nov 1, 2015

Current coverage is 100.00%

CarloLucibello commented Nov 3, 2015

jpfairbanks commented Nov 3, 2015

sbromberger Nov 3, 2015

Choose a reason for hiding this comment

CarloLucibello Nov 4, 2015

Choose a reason for hiding this comment

sbromberger Nov 4, 2015

Choose a reason for hiding this comment

CarloLucibello Nov 4, 2015

Choose a reason for hiding this comment

CarloLucibello commented Nov 4, 2015

sbromberger Nov 4, 2015

Choose a reason for hiding this comment

CarloLucibello Nov 5, 2015

Choose a reason for hiding this comment

CarloLucibello commented Nov 5, 2015

sbromberger commented Nov 5, 2015

sbromberger commented Nov 5, 2015

jpfairbanks commented Nov 6, 2015

sbromberger commented Nov 6, 2015

sbromberger commented Nov 11, 2015

CarloLucibello commented Nov 11, 2015

jpfairbanks commented Nov 12, 2015

sbromberger commented Nov 12, 2015

jpfairbanks commented Nov 13, 2015

Current coverage is `100.00%`