So the steps I should go by is to create a custom class or expose a class which makes use of RandomFourierDotFeatures and handles all the dependencies which I mentioned before, am I catching the drift? Once that is done I can go by creating a meta_example whose ported python example could serve as a reference for creating a notebook. This all sounds good!
I've one doubt though, youve mentioned " ore processors" I'm not quite sure what you meant by that. On Fri, Mar 27, 2020 at 2:37 PM Heiko Strathmann <heiko.strathm...@gmail.com> wrote: > Hi! > > (Always cc the public mailing list when discussing things, so others can > learn from this as well) > > You brought up an important point: the current Api of the Fourier features > might not be yet compatible with the new api. This is why you are facing > problems. This is actually a good thing, as updating the api is something > we want to do. For now, let’s make the existing rff architecture work with > the new api. > > I would suggest that you try to write a meta example for basic rff > embeddings. In order for that to work, you will need to add a new api base > class, or preferably expose the rff under an existing one. I think the ore > processors could be used for it, they do exactly what you described in your > email. It requires changing c++ code a bit. Have a look at merged prs > around meta examples and you will see that some of them tackle similar > issues for other classes. I can also have a look on the weekend. > > Once a meta example works, you can write the notebook. > > All this will help you understand the shogun codebase more and also help > with your application > > Let me know if you have any specific questions. > > H > > > On Thu, 26 Mar 2020 at 23:38, sai_ng <jonpsy...@gmail.com> wrote: > >> Hi Heiko, >> I'm currently trying to make a ipython-notebook for >> RandomFourierFeatures. I'm having a really hard time dealing with the >> specific data types which are needed to create "RandomFourierDotFeatures" >> type of feature, it asks for "DotFeatures" as input but I can't seem to >> find a way to create new DotFeatures or convert "shogun.Features" to that >> type. >> >> I think it would be better if it's all wrapped under a class and all >> these dependencies should've been encapsulated inside that class such that >> it takes "feature" input X and automatically returns subsampled >> "X_transform" much like how sklearn does. >> >> If we're going by to create the class, then I'm unsure as to what to >> write in my ipython-notebook as of now. I"d like clarification on this >> matter if possible, thanks in advance. >> > -- > Sent from my phone >
