January 2020 - graph-tool - archives.skewed.de

Re: [graph-tool] Stochastic Blockmodeling Result: Vertex and Edge colors
by Lietz, Haiko 21 Jan '20

21 Jan '20

Hi Paul, From how I understand things, you can get the used colors this way: from graph_tool.all import * num_blocks = 14 for i in range(0, num_blocks): j = int(i*(255/num_blocks-1)) print(default_cm(j)) Here ``num_blocks`` is the number of colors that are needed (in a celegans case: 14). Is this correct, Tiago or others? Haiko

2 1

Speeding up graph copy?
by Zouhair 21 Jan '20

21 Jan '20

Hi, I'm using graph-tool in my project, and my use case is as follows: 1) build an undirected graph with vertices and edges (~1e5 vtx and ~5e6 edges) 2) attach some properties to the vertices 3) for about ~5000 sets of *new* candidate vertices (each set has about 2-30 vertices) 3.a - add new vertices to graph 3.b - add edges as necessary between new vertices and old ones 3.c - update the properties 3.d - run an algorithm which depends on properties and edges, save result to a list 3.e - remove new vertices from graph and restore properties (i.e. undo 3.a-c) Steps 1) and 2) are going well. For step 3), the algorithm in step 3.d runs fast, 3.b and 3.c are also fast, however, the bottleneck I'm facing is 3.a and 3.e (about 80% of runtime is spent there) I've read through the docs and tried different approaches: i- make a copy of the graph using g.copy() or Graph(g): this avoids steps 3.e, but copying the graph is much slower ii - use a GraphView: adding vertices to the graph view adds them to the original graph, so still need to do step 3.e iii- use GraphView but only filter on the original vertices in the graph: this avoids having to delete the vertices, but making GraphView with a filter function is still slow (doing `gt.GraphView(g, lambda v: v < n_nodes_orig)` takes 75% of runtime, and the number of vertices in the original graph keeps growing... iv - do this using multiprocessing with the hope that pickling the object avoids having to copy the graph, but it it appears that a different process modifying the graph ends up affecting the graph that another process receives ( because multiprocessing doesn't pickle the underlying C++ object/memory?) In all cases, it's surprising that the bulk of the runtime is in making graph copies as opposed to running the algorithm. The cleanest solution is to take a copy of the graph and work on it to avoid having to later delete vertices and restore properties, so I'm hoping that there is a way to speed up graph copies? I'm hoping for some input on whether there is a better way to achieve this using graph-tool, or some pythonic magic to help speed up the copy operation. Thanks! P.S. sorry if this is a repost - my initial post was rejected by nabble, trying again directly.

2 1

Stochastic Blockmodeling Result: Vertex and Edge colors
by Paul Jung 21 Jan '20

21 Jan '20

Hello, I think I am having the same problem recently risen. I was just wondering how I can access to the color tuples of the given mapping (draw) results. Given that: >>> g = gt.collection.data["celegansneural"]>>> state = gt.minimize_nested_blockmodel_dl(g, deg_corr=True)>>> gt.draw_hierarchy(state, output="celegansneural_nested_mdl.pdf") I want to use the color codes of blocks (vertex) and edges for another mapping. Any ideas? Paul -- *Paul H. Jung* PhD Candidate in Geography & Urban Regional Analysis University of North Carolina at Charlotte 9201 University City Blvd, Charlotte, NC 28223 https://pages.uncc.edu/paul-jung --

2 1

Error with get_edges_prob in layered sbm
by Anatol Wegner 21 Jan '20

21 Jan '20

Hi, I am getting the following error when using get_edges_prob() with layered SBMs. Minimal example: import graph_tool.all as gt import numpy as np gr=gt.generate_sbm(b=np.array([0]*500+[1]*500),probs=np.array([[10000,200],[200,10000]])) etype=gr.new_edge_property('int') gr.ep.etype=etype for e in gr.edges(): gr.ep.etype[e]=np.random.choice([0,1,2,3]) state = gt.minimize_nested_blockmodel_dl(gr, deg_corr=True,layers=True,state_args=dict(ec=gr.ep.etype,layers=True),verbose=False) state.get_edges_prob([[2,32,0],[3,4,2]],spurious=[]) --------------------------------------------------------------------------- UnboundLocalError Traceback (most recent call last) <ipython-input-30-a1758ce2345d> in<module>() 7 gr.ep.etype[e]=np.random.choice([0,1,2,3]) 8 state= gt.inference.minimize_nested_blockmodel_dl(gr, deg_corr=True,layers=True,state_args=dict(ec=gr.ep.etype,layers=True),verbose=False) ----> 9state.get_edges_prob([[2,32,0],[3,4,2]],spurious=[]) /usr/lib/python2.7/dist-packages/graph_tool/inference/nested_blockmodel.py inget_edges_prob(self, missing, spurious, entropy_args) 481 lstate._state.clear_egroups() 482 --> 483L+= lstate.get_edges_prob(missing, spurious, entropy_args=eargs) 484 if isinstance(self.levels[0], LayeredBlockState): 485 missing= [(lstate.b[u], lstate.b[v], l_) for u, v, l_in missing] /usr/lib/python2.7/dist-packages/graph_tool/inference/layered_blockmodel.py inget_edges_prob(self, missing, spurious, entropy_args) 934 state= self.layer_states[l[0]] 935 state.g.remove_edge(e) --> 936for u, v, lin old_es: 937 if not l[1]: 938 state= self.agg_state UnboundLocalError: local variable 'old_es' referenced before assignment Regards, Anatol

2 1

Random seed in blockmodeling
by Lietz, Haiko 20 Jan '20

20 Jan '20

Hi all, Is it possible to use a random seed when calling minimize_blockmodel_dl() or minimize_nested_blockmodel_dl() such that the inferred state is always the same? Thx Haiko

1 1

Graph-Tool Windows-Usage
by PhilippCoelsch 17 Jan '20

17 Jan '20

Hi all, I am a PhD-student who wants to perform some network analyses using this Graph-Tool library. Unfortunately I only have access to a Windows 10 PC and no Mac OS or Linux device. I also cannot install another operating system on my computer. I tried the installation via Docker as described in the documentation, however, this is not working as Docker requires me to have an enterprise version of windows which I dont have either.. Furthermore, all articles on Nabble are from some years ago. So maybe there are some new updates to the windows-issue in the meantime? Would appreciate any help for setting up... Thanks in advance and kind regards, Philipp -- Sent from: http://main-discussion-list-for-the-graph-tool-project.982480.n3.nabble.com/

3 3

Accessing colors
by Lietz, Haiko 16 Jan '20

16 Jan '20

Hi all, How can I access the (r, g, b, alpha) tuples used by default when using graph_draw()? Thanks Haiko

2 3

Variable scoping for collect_vertex_marginals
by Davide Cittaro 16 Jan '20

16 Jan '20

Hi again, This is more a python related question, but I'm asking here hoping somebody had a similar issue and worked it out. I need to collect vertex marginals during equilibrate and when I do it in my python console everything works just fine (as in the cookbook): pv = [None] * len(state.get_levels()) def collect_marginals(s): global pv pv = [sl.collect_vertex_marginals(pv[l]) for l, sl in enumerate(s.get_levels())] gt.mcmc_equilibrate(state, force_niter=10000, mcmc_args=dict(niter=10), callback=collect_marginals) as far as understand, every time sl.collect_vertex_marginals is called, it takes as argument the histogram calculated in the previous iterations. My problem is that when I put this code into a function that is imported as a module, scoping is broken and I get an error like NameError: name 'pv' is not defined Shortly, I have the function nsbm in file _nsbm.py: def nsbm(…): […] pv = [None] * len(state.get_levels()) def collect_marginals(s): global pv pv = [sl.collect_vertex_marginals(pv[l]) for l, sl in enumerate(s.get_levels())] gt.mcmc_equilibrate(state, force_niter=10000, mcmc_args=dict(niter=10), callback=collect_marginals) which is imported from ._nsbm import nsbm by the __init__.py of the package module (named tools), then imported by the main from . import tools as tl the directory structure of the package is (among others) scanpy/tools/_nsbm.py scanpy/tools/__init__.py scanpy/__init__.py as far as I understand callback function passed to mcmc_equilibrate is not supposed to return anything, collect_vertex_marginals returns the histogram summing the current to the previous. I think I can't pass pv as an argument to the callback as it becomes local to the function. Of course, I'm now reading about variable scoping in python, but if you have any hint it would be greatly appreciated. Thanks d

2 2

About speed, epsilon and wait
by Davide Cittaro 16 Jan '20

16 Jan '20

Hi all, I'm trying to figure out a decent set of parameters to achieve a good trade-off between speed and accuracy. On my data, default parameters make comuputation times extremely long (especially compared to other community finding approaches), but we have noticed looking at the log that even 100 consecutive iterations without record breaking are quite rare, would it be reasonable to set wait=100, possibly increasing nbreaks, then? Also, are "wait" parameter and "niter" passed to mcmc_args somehow related, in the sense that it could be opportune to increase one and decrease the other or should be they considered independent? Reading the older posts I've found there's an "epsilon" parameter which could be tuned in order to converge faster, and we are definitely trying to increase it (1e-3, to start). I've noticed another epsilon in hierarchy_minimize() parameters, is that expected to influence speed as well? I may have additional questions on specific parameters (such as d and c) but I guess I can save them for another post. Thanks d

2 3

Postdominator
by Gerion Entrup 16 Jan '20

16 Jan '20

Hi, I have seen the algorithm for the dominator tree in graph_tool. Is there an algorithm present that calculates the postdominator tree [1] as well? Best, Gerion [1] https://en.wikipedia.org/wiki/Dominator_(graph_theory)#Postdominance

2 1