Replies: 7 comments 11 replies
-
|
Ok. I've confirmed that using So I'm now seeing if Can someone suggest an output format which was implemented to include the layer information? |
Beta Was this translation helpful? Give feedback.
-
|
After maybe 15 minutes, ---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
Input In [22], in <module>
1 module_layers_outpath_json = module_layers_outpath_pajek.with_suffix('.json')
----> 2 im.write_json(str(module_layers_outpath_json))
File /usr/local/lib64/python3.8/site-packages/infomap.py:6650, in Infomap.write_json(self, filename, states)
6636 def write_json(self, filename, states=False):
6637 """Write result to a JSON file.
6638
6639 See Also
(...)
6648 If the state nodes should be included. Default ``False``.
6649 """
-> 6650 return self.writeJsonTree(filename, states)
File /usr/local/lib64/python3.8/site-packages/infomap.py:3303, in InfomapBase.writeJsonTree(self, *args)
3302 def writeJsonTree(self, *args):
-> 3303 return _infomap.InfomapBase_writeJsonTree(self, *args)
RuntimeError: std::bad_allocIs that function going to need more than the ~160G of memory (and 3.6P of disk space) I have free? |
Beta Was this translation helpful? Give feedback.
-
|
Looks like I've got . . . > infomap --version
Infomap version 2.1.0 compiled with OpenMP
See www.mapequation.org for terms of use.I guess I'll see if I can get Pypi v 2.2.0, and see if any updates fixes things for me. |
Beta Was this translation helpful? Give feedback.
-
|
Hi and thanks for reaching out!
Why do you want to print the pajek file? It's not needed to run Infomap if you use the Python module. I'll move this to a discussion post and create a separate issue for the write_json bug for now. Please let us know if you have any questions. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the response, @antoneri , and thanks for your work on this open software! I was partly using the pajek output to confirm that the functions I was using to add layer information, were actually doing what I hoped/expected. What we are trying to do . . . If the layer information is not apparent in the pajek output, could you point me to the/a recommended way to see that? |
Beta Was this translation helpful? Give feedback.
-
|
Alright, well almost! Infomap represents multilayer networks using "state nodes". You can inspect the (relaxed) state network using something like this: >>> from infomap import Infomap
>>> im = Infomap(silent=True, no_infomap=True)
>>> im.add_multilayer_intra_link(1, 1, 2)
>>> im.add_multilayer_intra_link(2, 1, 3)
>>> im.run()
>>> im.write_state_network("example.net")
>>> !cat example.net
# v2.2.0
# ./Infomap --no-infomap --silent
*States
# stateId physicalId
0 1
1 2
2 3
3 1
*Links
0 1 0.925
0 2 0.075
1 0 1
2 3 1
3 1 0.075
3 2 0.925But my suggestion is to not focus so much on the internal representation, but instead study the "state tree" output file which includes the layer id when running on multilayer networks. Or, you can use the Python API: >>> from infomap import Infomap
>>> im = Infomap(silent=True)
>>> im.read_file("multilayer.net")
>>> im.run()
>>> for node in im.nodes:
... print(node.node_id, node.module_id, node.layer_id)
...
1 1 2
2 1 2
3 1 2
1 2 1
4 2 1
5 2 1We refer to clusters of nodes as "modules" (comes from the map equation using a modular description of a random walk on the network). Just so we avoid any confusion. |
Beta Was this translation helpful? Give feedback.
-
|
@antoneri , Thanks for taking the time to respond with all the useful information and examples! |
Beta Was this translation helpful? Give feedback.
-
|
I don't know if this is the best place (topic, under which) to continue asking things. Anyway, I was a little hopeful and surprised to see the 'preferred_number_of_modules' parameter, as a possible easy way to get to our target. But when I set that to 40, for example, I see things like . . . |
Beta Was this translation helpful? Give feedback.
-
|
Thanks so much for the response, @antoneri I create a link for every one. If both of any of those participate in any one of ~5 biological "maps", that link is an intra-layer link. I create no explicit inter-layer links. But I guess, implicitly, all other links will be inter-layer. I'm also wondering . . . |
Beta Was this translation helpful? Give feedback.
-
|
Just to clarify, by modeling I mean if you should represent your data as a regular "single layer" or multi-layer network, what the weights should be, what it means for a random walker to transition between layers, and how likely that should be. And so on... :) As you might already know, Infomap finds flow modules that trap a random walk for a relatively long time. This is why network representation matters so much for community detection. Since Infomap 2.0, we include self-links by default. You can omit them with I'm not familiar with all your terms, so I can only help with what I think I recognize. I'm however of the opinion that it's best to get some results with the simplest possible model to verify that there's any meaningful community structure in your data. |
Beta Was this translation helpful? Give feedback.
-
|
Am I wrong in assuming I can induce communities, to test to see if I can get any modules (not a higher number than my total number of links), by increasing the weights on certain links? im.run(
include_self_links=False,
preferred_number_of_modules=40,
multilayer_relax_rate = 0.15,
)Here's my run . . . I only created 14091 links. In this case I made the link weights of interest 9. |
Beta Was this translation helpful? Give feedback.
-
|
Now I think I know what's happening! Are you using I also see this line:
Something is wrong if you added 14000 links. Maybe the same cause as above. How many nodes/layers are you adding? |
Beta Was this translation helpful? Give feedback.
-
|
I realize I actually am still using Thanks, @antoneri ! Also, while I'm here, I've been considering trying NetworkX. It looks like that has some handy metadata, and adjacency, functionality, we might take advantage of. I also like that they can use strings (any hashable type) for node ids. Do you have any experience with that software. I'm not sure if, maybe, it is mostly pure Python, and might not be fast enough for large networks. I wonder about export/import to Infomap or better visualization tools. Ah . . . I just noticed |
Beta Was this translation helpful? Give feedback.
-
|
This looks much better to me! I actually kept I've been reading the docs, but I'm not completely clear on how to show those top "(30 non-trivial) modules". Also, how might I filter for only the modules which include at least one of our nodes of interest? Thanks so much for your time! |
Beta Was this translation helpful? Give feedback.
-
|
It would be best if you did not use This is why you get this warning:
I'm telling you: don't do it. Your results are wrong if you do. You can get the modules using |
Beta Was this translation helpful? Give feedback.
-
|
Ah . . . Gotcha. So there is more than one issue. Yeah . . . I was mainly using it, assuming the graph would be most authoritative in terms of what node ids were already in use. With previous ones added, I figured |
Beta Was this translation helpful? Give feedback.
-
|
I removed But it does look like I might be now getting the 40 modules I requested. I only use . . . Do you think I should replace |
Beta Was this translation helpful? Give feedback.
-
I've created a map, and done a bunch of . . .
add_multilayer_intra_linkcalls.But
write_pajekincludes no "*Intra" section (nor any Multilayer additional information, at all.)If that exporter doesn't support that, can you point out which one does?
Maybe
write_json?I guess I can test and see . . .
Perhaps I am just misunderstanding the scope/capabilities of infomap.
Would it work if I just used
add_multilayer_link, but with the same layer id?I'm wondering if since "Multilayer" is not optional in the pajek spec, if the Infomap object might not even initialize that member/attribute if something with 'multilayer' in the function name is never called.
Basically I have some semantic information which can allow grouping Vertices into layers. I was hoping to use the Infomap functions to integrate this with some binary relationship information (e.g. which can be used to create edges).
Beta Was this translation helpful? Give feedback.
All reactions