neurodata / SPORF

This is the implementation of Sparse Projection Oblique Randomer Forest
https://neurodata.io/forests/
97 stars 46 forks source link

[WIP] Add two feature sampling methods for graphs #337

Open j1c opened 5 years ago

j1c commented 5 years ago

These are two methods for graph valued data.

  1. Graph-Edge-MORF randomly takes one row of the adjacency matrices, then randomly samples some number of elements of that row based on patch_height_min and patch_height_max. The sum of the selected elements is the new feature for that graph.

  2. Graph-Node-MORF randomly samples n \in [patch_height_min, patch_height_max] integers between [1, image_height]. You take the rows and cols based on the sampled integers, and sum the values in that small square. This is the new feature for that graph.

netlify[bot] commented 5 years ago

Deploy preview for rerf ready!

Built with commit b5839d10097adeca031ac529a14b6395049005a9

https://deploy-preview-337--rerf.netlify.com

j1c commented 5 years ago

Not quite sure how to deal with different sized tabs cause tabs in C++ files vary in size. Can be 8 spaces or 4 or 2....