Is Graphlab updated with cxxnet V2?

User 1738 | 4/9/2015, 9:09:07 AM

Hi, I have two questions please: 1. Have you updated graphlab with cxxnet v2 for deep learning?. I believe it now supports online augmentation and multi-gpu which is important 2. I know SFrame can extend to a size larger than the permissible ram limit. If I grow a random forest or svm on SFrame data will it be able to deal with size of data larger than ram limit?

thank you very much

Comments

User 1190 | 4/9/2015, 5:18:24 PM

Hi @raphael,

It's on our road map to integrate with cxxnet v2 and support multigpu and more flexible network architectures. Please stay tuned.

For your second question, yes. But keep in mind some algorithms are more memory intensive than the others, e.g. using sgd solver for logistic regression on high dimension data versus using newton's method is going to have different scalability limit. We are working hard to push the limit for all our algorithms, so they scale beyond the limit that most people are facing with in-memory only implementations.

Thanks for your question. -jay