ValueError in Network Demo

[Fabian Dablander]

Hi there,

Not sure what I am doing wrong. I am trying to run the network demo locally, but I get the following error:

The error persists for scale values up to and including 19, and goes away for values larger than 19. Any ideas what is happening?

Thanks!

[Ryan Seamus McGee]

Thanks for reaching out, I'm sorry you're encountering this issue. I just tested the notebook on my machine and unfortunately I'm not able to reproduce this error on my end. It appears that in your case baseGraph has one or more nodes that have no edges which causes neighbors to be an empty list that numpy.random.choice can't handle. Are you using the baseGraph generation that was originally in the notebook, or did you modify how that graph is created? The current implementation of custom_exponential_graph implicitly assumes that all nodes have non-zero degree, but I will change the code to be safe for cases where nodes do have zero degree. I'll post here when that is updated on the repo and pip. If you think that is not the source of your error let me know!

[Fabian Dablander]

Thanks for the quick response! Hmm, it is odd that you cannot reproduce it. I did not change anything in the notebook, but ran it as is. I don't think the explanation you suggested is correct, because

0 in dict(baseGraph.degree).values() # False

returns False, which means that all nodes have non-zero degree, right?

[pjasc]

I am having the exact same error..

[Ryan Seamus McGee]

Thank you both for re-elevating this issue. I know a bunch of people have been able to run the notebooks and package generally, so the problem here seems to be a strange one, but I will investigate further. If you can let me know what version of python you're running and any other environment factors that might be relevant that'd be helpful.

[pjasc]

I have python 2.7.15. Basically I have downloaded anaconda with python 2.7 and then launched Jupiter Notebook and ran your code

[pjasc]

I created a virtual environment with python 3.7 and everything works fine now! So I guess there was some kind of issue with the python version

[Ryan Seamus McGee]

Thanks @pjasc, that's helpful for me to know in tracking down the issue for python 2 users.

[Adam Tauman Kalai]

I'm having this issue with Python 3.6.8 and numpy version 1.15.1
Should I try python 3.7?

[Adam Tauman Kalai]

I found the problem. You can either upgrade to a newer version of numpy, or better yet the code could be modified not to sample from an empty list. See the comparison below:

import numpy as np
print(np.version)
np.random.choice([], size=0, replace=False)
1.16.2
array([], dtype=float64)

but in another venv:

import numpy as np
print(np.version)
np.random.choice([], size=0, replace=False)
1.15.1

---

ValueError Traceback (most recent call last)
in
1 import numpy as np
2 print(np.version)
----> 3 np.random.choice([], size=0, replace=False)

mtrand.pyx in mtrand.RandomState.choice()

ValueError: a must be non-empty

[Fabian Dablander]

Excellent, upgrading numpy fixed it for me; but I agree that it would be best to update the code. Thank you!