search

corail-research / learning-generic-csp

Representing CP problems as graphs
16 stars 1 forks source link

More Data #38

Open x-lad opened 6 months ago

x-lad commented 6 months ago

Hi I wanted to know if it's possible to upload your datasets? For example, instance_generator.py of Graph Coloring takes a very long time - which is unrealistic.

Additionally, could you upload your pre-trained weights?

Thanks

jardinetsouffleton commented 6 months ago

Hi Elad! Thanks for reaching out. For the datasets; unfortunately they are not available anymore as they took too much space. As for the pre-trained weights; that's a mistake on my end; I'll upload them over the course of the next days.

x-lad commented 6 months ago

Thanks for your fast response.

I'm looking forward for the pre-trained weights - thanks!

Regarding the dataset, do you have any two cents about why it's taking so long? It seems like the solver takes forever in some instances (n>50, c>=6). Did you just let it run for a long time? May I ask which CPU you used (GPU is mentioned in the paper)? If you have any ideas/suggestions on what I can do, it would be much appreciated.

Thanks again

jardinetsouffleton commented 6 months ago

First the solver is slow for this type of instances, so switching to gurobi/cplex if possible would make for a huge speedup. Next, on top of solving the problem a first time, the process then adds one edge at a time and re-solves the instance so that only instances where the k-colorability changes when we add one single edge. So it can take a lot of time...

jardinetsouffleton commented 6 months ago

As for the CPU, i am not sure

x-lad commented 5 months ago

I created the dataset Graph Coloring. If you want, I can share it with you to go along with your code