GSoC_2017_applications_webdemos
(please read the general description of application projects here)
A cute stand-alone GSoC project could be to pimp our web-demos, like binary classifier, GP regression, hand-written digit recognition. These existing ones are nice, but a bit boring -- so let's push them to the next level.
- Heiko (github: karlnapf, IRC: HeikoS)
- Viktor (github: vigsterkr, IRC: wiking)
- Lea (github: lgoetz, IRC: leagoetz)
The project would basically involve going crazy (and we mean it) with all the cool new visualisation frameworks out there, be creative with real-world datasets to be used in there, and creating a nice website to showcase the thing. (Such as this mobility data visualisation - except your project would of course be showing off Shogun's algorithm on data, not just the data itself).
Some things we would like to see involve
- Density estimation / spatial modelling for interactive maps
- Embedding Shogun in a larger pipeline of tools
- Making the existing interactive demos more fancy, pretty, and cool.
- Creative usage of Shogun's algorithms: rather than composing together points for regression, why not take a photo of the user with the webcam, then classify their gender/age/happiness? (we have done such things with Shogun in the past)
Please note that this project really requires a lot of self motivation and initial brainstorming from your side. We won't tell you exactly what to do (we will share our ideas with you though :) ).
For your application, the most important thing is that you have a clear vision of what you want to do. To develop that, start by running all the existing web demos locally and try to come up with cool ideas of how to improve them. As Shogun is a lot about education around ML, you could go in that direction. In fact, check out many of the notebooks we have. Some of the plots in there are really cool and it would be nice to have them in an interactive manner
- As this project is featuring your ideas, the first step is to show them to us: create a mock-up of what you have in mind.
- The next step would be to pin down the technology to use
- The next step (a good entrance task) is to build a simple proof of concept of what you have in mind
- Then, choose exactly which algorithms / pipeline you want to create a demo for
Entrance tasks:
- fix issues in the existing demos (we did not test them in a while),
- create a Docker container (on Dockerhub) to fire up the demos
- add a new visualisation idea
Some ideas for new visualizations:
- explain the effect of regularisation in regression
- visualise support vectors in the SVM classifier
- model selection and how different parameters affect the model
While the project is meant to be stand-alone, we would be more than happy to integrate any cool outcomes in Shogun itself, i.e. run your demos on our servers. Definitely the outcome should be a self contained (maintainable) Docker container that can be used to run the demos anywhere.
Because visualisations are cool! And because you have the freedom to really go crazy with your favourite technologies. Currently, the demos are very minimal and simple. It would be cool if they would cause some "ah" effect. That can be achieved with cool unusual ideas how to visualise things, with very fancy visualisation skills (using cool libraries to do that), and with a slick polished look (currently the demos look a bit ugly).
You name them!