Hamming Distance

[jac2130]

I added a file that calculates the simple Hamming Distance between two (possibly directed) graphs.

[hagberg]

Maybe it is simpler and faster to not use numpy? e.g. (untested)

def hd(G, H):
    count = 0
    for e in G.edges_iter():
        if not H.has_edge(*e):
            count+=1
    for e in H.edges_iter():
        if not G.has_edge(*e):
            count+=1
    return count

[jac2130]

That is probably better and it would equally work; sure! Actually, you
might have an idea about how to compute a "Generalized Hamming Distance"
which is what I happen to need for a problem that I'm working on: I have an
alphabet of 8 types of relations (positive, negative, zero, non-zero,
non-positive, non-negative, non-zero, missing) and I need that the
differences between members of the alphabet are ordered as follows:

d(positive, negative)>d(positive, non-positive)=d(negative,
non-negative)>d(non-positive, non-negative)=d(non-positive,
non-zero)=d(non-negative, non-zero)>d(positive, missing)=d(negative,
missing)>d(non-negative, missing)=d(non-positive, missing) >d(positive,
zero)=d(negative, zero)>d(non-negative, zero)=d(non-positive, zero)
....etc.

Any ideas?

JC

On Wed, Jun 26, 2013 at 11:58 AM, Aric Hagberg notifications@github.comwrote:

Maybe it is simpler and faster to not use numpy? e.g. (untested)

def hd(G, H):
count = 0
for e in G.edges_iter():
if not H.has_edge(_e):
count+=1
for e in H.edges_iter():
if not G.has_edge(_e):
count+=1
return count

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/893#issuecomment-20067337
.

Johannes http://www.columbia.edu/~jac2130

---

"I can calculate the motions of the heavenly bodies, but not the madness of
people."

• Isaac Newton

[dschult]

I made a pull request to your pull request with some format changes and file placement changes.
jac2130#1

Just for completeness the text of that pull request comment is:

Hi JC,
I'm reviewing your pull request to NetworkX with Hamming distance functions. I've moved the file into the algorithms folder and changed the formatting of the doc_string for your first function. Could you update the doc_strings into a similar format for the other functions? Also I added some lines at the top--see if you are OK with the more verbose names I proposed there. Its the networkx style not to abbreviate in function names. When the doc_strings are updated you can push back to your master branch and it will update the pull request too. Let me know if git causes trouble.

Finally, we will need some simple tests before merging. If you can come up with some simple graphs and compute correct values for them that's a good start--and maybe all the testing we need at first. You can create a test_hamming.py file in the tests directory inside algorithms (use test_cycles.py as an example if you want). Or if that's too much, just put the code for the tests in your hamming.py file after a line like:

if name == "main":
#test_code here

Thanks!

[jac2130]

Dear networkx team, I have added a class named BayesNet, which inherits from your DiGraph class, as well as from Allan Downey's Joint class. This class is meant to represent a Bayes Net, which is both, a directed graph with causal interpretation of the directed edges and a joint distribution. Although, right now the code is far from ideal, I hope that it may be of use to you and others!

[CrazyPython]

@jac2130 even if they did accept your code, the checks have failed.

[MridulS]

I'll go ahead and close this PR, this hasn't been touched in a long time and the code bits from BayesNet can't go in NetworkX as it's GPL.

@jac2130 I have created a new draft PR #5653 with only the new distances and made some changes to make it work with main branch of NX (it's not really tested) but feel free to pick it up from there if you interested in getting this in NetworkX 🎉