Skip to content

Commit

Permalink
Clean up RFFGaussPreproc (#5010)
Browse files Browse the repository at this point in the history
  • Loading branch information
@jonpsy
jonpsy authored Jun 10, 2020
1 parent ab6a696 commit 9158d3f
Show file tree
Hide file tree
Showing 4 changed files with 204 additions and 552 deletions.
2 changes: 1 addition & 1 deletion data
Submodule data updated 1 files
+1 −1 testsuite/meta/preprocessor/random_fourier_features.dat
Loading

2 comments on commit 9158d3f

@karlnapf
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@jonpsy so for performance I would check the apply methods for vector and matrix, and the fit/train method.
I would do this num_rff_feats = 100, 1000, and a input_dim = 10, 100, num_input_vectors = 100, 1000 (for apply matrix only, for apply vector you only need a single one). These will be 8 checks for apply_matrix, 4 for apply_vector, and 8 for train/fit

You should write those in c++, and put a file next to the files that you changed above. See e.g. here how to create benchmarks (you can also search for other benchmarks in the repo to learn more)

you can activate benchmark building using this switch.

and if you want to add a new benchmark you have to tell cmake to compile it, see e.g. here

Also you will see that there already is a benchmark that is related:https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/features/RandomFourierDotFeatures_benchmark.cc
I suggest you clean that one up, maybe move it, and use it to add your own.

@jonpsy
Copy link
Contributor Author

@jonpsy jonpsy commented on 9158d3f Jun 17, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the detailed information, I'll get into it.

Please sign in to comment.