search

dinhinfotech / DiGI

3 stars 1 forks source link

What are the memory requirements of DiGI? Why can't 256g of memory successfully run the data of cross_validation_2? #1

Open Spirit-0814 opened 2 years ago

Spirit-0814 commented 2 years ago

RT

dinhinfotech commented 2 years ago

The click decomposition has high complexity if the graph is dense and has a high number of cliques. So to reduce the memory consumption you can disable the clique decomposition or set a high threshold.

Spirit-0814 commented 2 years ago

Thank you for your reply. The data I use is the HumanNet_CF and HumanNet_PI you provided. I wonder if you keep the kernel_matrices generated by these two networks. I really need it.

dinhinfotech commented 2 years ago

I cannot check it now since I am on my vacation. I can do it and tell you later when I am at work.

Spirit-0814 commented 2 years ago

I'm sorry to disturb your vacation. When I set the clique thresholds to 50, it can run successfully, but the results on my Alzheimer's disease data set are not very good, AUC is only 0.69. Will setting clique thresholds too large cause model degradation?