Roadmap Shogun 2015 hack
Clone this wiki locally
Shogun hack 2015 Ideas
This is a list of topics to address when all Shogun developers get together. We aim to have a structured list of items, with links to existing or newly created github issues. Feel free to merge this with content from the other Wiki pages or discussions.
- A meta-langauge for Shogun example: #2508.
- Website updates (Kevin had some ideas?)
- Developer information (wiki or website?) #2533
- Manifest "What is Shogun? What tries Shogun to be?"
- Notebooks of things that are not yet covered
- Binary packages for various OS
- Improve Shogun development guidelines
- Scheduling a separate release for "easier installation": packages and scripts for Ubuntu/Debian/MacOSC https://github.com/shogun-toolbox/shogun/milestones/Shogun%203.4.1
Efficiency & Clean-ups
- Efficient and flexible data structure for n-dimensional array #2123 #2530
- Modularise Shogun to reduce compile memory requirements #2437 (some prototype is https://github.com/lisitsyn/aer)
- Serialisation via external library (Viktor?)
- General polishing of base classes #2113(Thoralf?)
- DPointers (again https://github.com/lisitsyn/aer)
- Parallelism backend interface
- Populate internal Linear algebra interface
- Stan for autodiff and MCMC #1875 #1998 #1929
- Vowpal Wabbit update
SHOGUN goes large-scale
- Investigate how non-toy examples scale in SHOGUN.
- Increase awareness of benchmarking and profiling:
- Identify hotspots using selected examples and right tools (gprof?).
- Create minimal example(s!) which covers hotspot and benchmark.
- Start optimizing minimal example(s!) and benchmark again.
- Developers should know how to benchmark and profile: Small document to describe this?
- If you know examples that scale better in other frameworks, please document them here, so we can investigate:
- Better support of sparse-datastructures and online-algorithms.
- Interface with MOE for better model selection