-
-
Notifications
You must be signed in to change notification settings - Fork 1k
GSoC_2017_Amalgamation
Sergey Lisitsyn edited this page Mar 1, 2017
·
1 revision
- [Sergey](Sergey Lisitsyn)
Medium to hard. It would require good understanding of libraries and different languages. C++ and JavaScript would be important skills.
The name of this project comes from amalgmation in the mxnet library, the process of combining whole model into a module with a few to zero dependencies.
In python this could work like
import pretrained_model
pretrained_model.apply(features)
in JS it could be something even more dynamic
fetch('/model/1').then(...).then(model => model.apply(features));
More description to come.
No devops likes to manage dependencies. Making Shogun able to compile models into simple modules that require no deps would be a way into real production systems.