Reproducible results

User 1583 | 3/21/2015, 8:02:26 AM

Hello there. I participating kaggle competition and have question about reproducing model results. In python sklearn package I can set random.seed() to init value and model results would be exactly the same, but its not true for graphlab packages (donno about all, but boostedtreesclassifier in my case). I guess, its due to C++ core of all algorithms, but does any way to reproduce graphlab's ML algorithm results?

Comments

User 91 | 3/22/2015, 4:10:54 AM

We try our best to make sure the models are reproducible. It looks like we may have missed out on the same using boosted trees. We will definitely take this as a feature request and add the random seed parameter to it.


User 1583 | 3/22/2015, 3:05:13 PM

Thank you!