jdmoorman / uclasm Goto Github PK

The name uclasm is inappropriate since it has UCLA in it and does not related directly to the topic of the code.

Precompute the cost of assigning each template edge to each world edge and store these things efficiently so we don't ever have to recompute them.

It would be nice for usability purposes to have a from_networkx function since networkx is the gold standard in network datastructures.

Thank you for sharing this work. Could you share an example and benchmarks for feature wise subgraph matching ?

The constrained rectangular assignment problem solvers will be split off into their own repository. Once that repository is up and running, they will disappear from uclasm.

The current implementation of count_alldiff on the master branch is gross and all of the names are terrible. Start by renaming all of the variables and consider reimplementing from scratch. Take advantage of the notion of "equivalence classes".

In the noiseless case, one of our many bottlenecks is counting the solution to the alldiff problem. We can probably do this in numba, so if that's a possibility we should try it since it will likely be a lot faster.

This is one of many bottlenecks. We should be able to use numba for this when we get around to doing the precomputed edge costs.

MatchingProblem should keep track of which template nodes have exactly one candidate.

I try to run the project, but got some errors.

AttributeError: module 'uclasm' has no attribute 'nodewise_cost_bound'

Is the tutorial correct?