computing full matrix for radial maps with hub provided about lomap

[arjun-nufc]

Nevermind, I think that this is related to issue #24, and that explains why you still need to compute the full matrix of similarities. I am looking for a simple hub and spoke radial graph. it looks like this is possible with cutoff=0.0 and allow_tree=True.

[arjun-nufc]

This is still an issue though - when trying to explore larger numbers of synthetic possibilities in a real world application, a simple hub and spoke radial graph is going to be needed. While this is possible as mentioned above, it still requires computation of the full Nx(N-1)/2 similarities. This gets to be prohibitively slow pretty quickly.

[IAlibay]

@arjun-nufc I believe that the functionality you need is present in the openfe toolkit. If you want to raise an issue there with the task you're trying to do I'm pretty sure we can give you a code snippet that'll achieve a radial network with the lomap mapper & scorer in a minimum number of evaluations.

[richardjgowers]

@arjun-nufc this gist shows how to create a radial network using openfe and lomap together: https://gist.github.com/richardjgowers/79451ed3d698618125f9fa1b3dd82e52 This will only require the N-1 calculation of MCS/score as expected.

Making the traditional Lomap GraphGen class not require the full N(N-1)/2 calculations would require a lot of rewriting that isn't currently planned so I'm going to close this issue