Fix GraphMatch random_state behavior with n_init > 1, parallelize

[asaadeldin11]

• Does this PR add any new dependencies?
• Does this PR modify any existing APIs?
  • Is the change to the API backwards compatible?
• Have you built the documentation (reference and/or tutorial) and verified the generated documentation is appropriate?

Reference Issues/PRs

Closes #761
Closes #765.

What does this implement/fix? Briefly explain your changes.

Fixes bug where random_state when n_init > 1 was always the same. Also parallelizes over inits

[netlify]

Deploy preview for graspologic ready!

Built with commit 6c5c27c

https://deploy-preview-770--graspologic.netlify.app

[bdpedigo]

added a test, feel free to change it

[bdpedigo]

Think this is good to go, @asaadeldin11 you can decide whether to just leave a comment about efficiency or see if the n_jobs = 1 fix helps performance, otherwise I'm ready to merge

[bdpedigo]

@asaadeldin11 somehow that broke old tests, any idea why? maybe the tests just werent that stable to initialization? Or do we think something else is going on?

[asaadeldin11]

@asaadeldin11 somehow that broke old tests, any idea why? maybe the tests just werent that stable to initialization? Or do we think something else is going on?

Strange, they were passing for me locally. The padded sgm test makes sense, i remove two nodes and get 48/50 match ratio, so I'll change the 1.0 to 0.95 since there's no reason to expect the isolates to match correctly.
The second one I'm less sure about, but I think it might be resolved by changed n in test_sim from 90 to something larger (120 or 150). Thoughts?

[bdpedigo]

@asaadeldin11 somehow that broke old tests, any idea why? maybe the tests just werent that stable to initialization? Or do we think something else is going on?

Strange, they were passing for me locally. The padded sgm test makes sense, i remove two nodes and get 48/50 match ratio, so I'll change the 1.0 to 0.95 since there's no reason to expect the isolates to match correctly.
The second one I'm less sure about, but I think it might be resolved by changed n in test_sim from 90 to something larger (120 or 150). Thoughts?

is there any other way to just make the problem easier?

[asaadeldin11]

is there any other way to just make the problem easier?

I can just make the similarity matrix the identity matrix, and make it an ER graph?