Add non-backtracking spectral distance

[leotrs]

Pioneered by one Leonardo Torres linky.

Apologies for the mess of commits...

[sdmccabe]

ImportError: No module named 'ot'

[tlarock]

#133 is included in this PR.....

Edit: Looks like you changed it back, thx

[leotrs]

@sdmccabe do we want to add ot to the general requirements file? If we're ever thinking of shipping this, it might be nice to keep the global requirements to a minimum. We can fix the tests using a special requirements_test.txt file which includes everything.

[tlarock]

A bit concerned that it seems like the POT library (good name) has not been actively maintained since fall 2018 (https:/rflamary/POT/pulse/monthly), which could indicate that the maintainers are moving on. Are there any other EMD libraries out there we could use?

[sdmccabe]

I see two approaches:

1. Add ot to general requirements. I'm fine with this.
2. Have a requirements_test.txt file. This would be ok, but I have two concerns. One: making sure what we're testing and what we're using doesn't wind up too far out of alignment. Two: If we do this, we should explicitly handle the importerror, tell users that it's not available without this dependency, and include that information in the docstring.

[leotrs]

@tlarock I see what you're saying but (i) Fall 2018 is not too bad (and I got response from them as early as Jan 10 here, and (ii) I don't think this is an immediate problem. The scenario where our code breaks because this one dependency is outdated seems very far away, maybe years? And even in that case, the only part of our code that breaks is this one distance (which is not even a good one imho...)

@sdmccabe Agreed. I don't think it's too hard to make sure that we are always testing more than what we are using, precisely because requirements_test.txt would contain more imports than requirements.txt. However, you raise a good point about signaling in some way that there are imports particular to distance methods. The simplest solution is to add POT (good name) to requirements.txt and only have a requirements_test.txt if this same problem arises more times in the future.

[sdmccabe]

No one's objected, so I think we can just add POT to requirements.txt and get this merged.

[sdmccabe]

3.82s$ pip install -r requirements.txt
Collecting networkx>=2.2.0 (from -r requirements.txt (line 1))
  Downloading https://files.pythonhosted.org/packages/f3/f4/7e20ef40b118478191cec0b58c3192f822cace858c19505c7670961b76b2/networkx-2.2.zip (1.7MB)
Requirement already satisfied: numpy>=1.10.0 in /home/travis/virtualenv/python3.5.6/lib/python3.5/site-packages (from -r requirements.txt (line 2)) (1.15.0)
Collecting scipy>=1.0.0 (from -r requirements.txt (line 3))
  Downloading https://files.pythonhosted.org/packages/f0/30/526bee2ce18c066f9ff13ba89603f6c2b96c9fd406b57a21a7ba14bf5679/scipy-1.2.1-cp35-cp35m-manylinux1_x86_64.whl (24.7MB)
Collecting scikit-learn>=0.18.2 (from -r requirements.txt (line 4))
  Downloading https://files.pythonhosted.org/packages/38/7f/7d21bd89a97603a14d3770ca912d4402c394a8e50642f9b8b5cc6867a61a/scikit_learn-0.20.3-cp35-cp35m-manylinux1_x86_64.whl (5.3MB)
Collecting POT==0.5.1 (from -r requirements.txt (line 5))
  Downloading https://files.pythonhosted.org/packages/28/4b/7aaa1f840a359f5953dd378e0237fa8faf9b0a415ff7282b7375fbe68d27/POT-0.5.1.tar.gz (720kB)
    Complete output from command python setup.py egg_info:
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/tmp/pip-install-4rw5e1vw/POT/setup.py", line 7, in <module>
        from Cython.Build import cythonize
    ImportError: No module named 'Cython'

[leotrs]

This looks related to this issue. I'll try installing Cython beforehand.

[sdmccabe]

Out of curiosity, why == instead of >=? I just want to get this merged already so we can play with it, so don't change it, but just wondering.

[sdmccabe]

$ pip install -r requirements.txt
Collecting networkx>=2.2.0 (from -r requirements.txt (line 1))
  Downloading https://files.pythonhosted.org/packages/f3/f4/7e20ef40b118478191cec0b58c3192f822cace858c19505c7670961b76b2/networkx-2.2.zip (1.7MB)
    100% |████████████████████████████████| 1.7MB 873kB/s 
Requirement already satisfied: numpy>=1.10.0 in /home/travis/virtualenv/python3.6.3/lib/python3.6/site-packages (from -r requirements.txt (line 2))
Collecting scipy>=1.0.0 (from -r requirements.txt (line 3))
  Downloading https://files.pythonhosted.org/packages/7f/5f/c48860704092933bf1c4c1574a8de1ffd16bf4fde8bab190d747598844b2/scipy-1.2.1-cp36-cp36m-manylinux1_x86_64.whl (24.8MB)
    100% |████████████████████████████████| 24.8MB 60kB/s 
Collecting scikit-learn>=0.18.2 (from -r requirements.txt (line 4))
  Downloading https://files.pythonhosted.org/packages/5e/82/c0de5839d613b82bddd088599ac0bbfbbbcbd8ca470680658352d2c435bd/scikit_learn-0.20.3-cp36-cp36m-manylinux1_x86_64.whl (5.4MB)
    100% |████████████████████████████████| 5.4MB 288kB/s 
Collecting Cython==0.29.6 (from -r requirements.txt (line 5))
  Downloading https://files.pythonhosted.org/packages/e1/fd/711507fa396064bf716493861d6955af45369d2c470548e34af20b79d4d4/Cython-0.29.6-cp36-cp36m-manylinux1_x86_64.whl (2.1MB)
    100% |████████████████████████████████| 2.1MB 737kB/s 
Collecting POT==0.5.1 (from -r requirements.txt (line 6))
  Downloading https://files.pythonhosted.org/packages/28/4b/7aaa1f840a359f5953dd378e0237fa8faf9b0a415ff7282b7375fbe68d27/POT-0.5.1.tar.gz (720kB)
    100% |████████████████████████████████| 727kB 1.9MB/s 
    Complete output from command python setup.py egg_info:
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/tmp/pip-build-93h2l13g/POT/setup.py", line 7, in <module>
        from Cython.Build import cythonize
    ModuleNotFoundError: No module named 'Cython'

Yikes.

[leotrs]

Because I copy pasted the output of `pip freeze`. It should probably be `>=`.

…

On Mon, Mar 11, 2019 at 3:37 PM Stefan McCabe ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In requirements.txt <#134 (comment)>: > @@ -2,4 +2,5 @@ networkx>=2.2.0 numpy>=1.10.0 scipy>=1.0.0 scikit-learn>=0.18.2 +Cython==0.29.6 Out of curiosity, why == instead of >=? I just want to get this merged already so we can play with it, so don't change it, but just wondering. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#134 (review)>, or mute the thread <https:/notifications/unsubscribe-auth/ABC8AODPpTa2fWGAUKU1murji41XYFqLks5vVrB7gaJpZM4bfsE4> .

-- www.leotrs.com | [email protected] PhD student at the Network Science Institute, Northeastern University

[sdmccabe]

I'll note that the POT project has a bare cython dependency in their requirements.txt and that one of the other projects referenced in that issue mentions having trouble with cython 0.29.

[sdmccabe]

    def test_same_graph():
        """The distance between two equal graphs must be zero."""
        G = nx.karate_club_graph()
    
        for obj in distance.__dict__.values():
            if isinstance(obj, type) and BaseDistance in obj.__bases__:
                dist = obj().dist(G, G)
>               assert dist == 0.0
E               assert 0.02209708691208376 == 0.0
test_distance.py:22: AssertionError

[sdmccabe]

The stochasticity is going to keep that test from passing. For now let's except it from the test (and note it in the docstring).

The test can be modified to something like the following:

def test_same_graph():
    """The distance between two equal graphs must be zero."""
    G = nx.karate_club_graph()

    for label, obj in distance.__dict__.items():
        if label in [
                'NBD'
        ]:
            continue
        if isinstance(obj, type) and BaseDistance in obj.__bases__:
            dist = obj().dist(G, G)
            assert dist == 0.0

We'll probably need to do the same for the symmetry test.

[sdmccabe]

We'll need to update the docs, but let's get this merged.

[leotrs]

I'll leave this here for posterity. POT depends on Cython which depends on python3.6-dev (or python3.5-dev). In case we have to deal with this again in the future.