napetrov

3 ms·

1. 1.▲
   by napetrov 5y ago

   yes, notebooks have no difference for sklernex You likely not exporting algorithms from scikit learn again after patching - i.e. patch call should be made prior to import. Note there are kaggle notebooks that showcase same optimizations -
2. 2.▲
   by napetrov 5y ago

   vectorization, threading, cache optimizations, etc...
3. 3.▲
   by napetrov 5y ago

   The point is that sklernex would bring performance for all X86 architectures, not just Intel. And yes scikit developers already working on generic improvements there
4. 4.▲
   by napetrov 5y ago

   patching is just one of options to enable optimizations - You also can do already sklearnex import SVC There are other ways how you can enable this https://intel.github.io/scikit-learn-intelex/what-is-patchin...
5. 5.▲
   by napetrov 5y ago

   You have to import KMeans again after patching. Also in you example it's tiny problem size that have lot of fluctuations. Basically in you code you are running stock both times. this extension also would bring perf to AMD, although Int