GSoC_2017_applications
This year, we would like to try something new: using Shogun on “real world” data to demonstrate its capabilities beyond recognising handwritten digits or Iris species. In practice, this will involve a mix of using and modifying Shogun. The idea is that the projects are stand-alone and the result is something really cool.
Important: While these projects have an analysis part where you try to figure out what is going on in your data, the major part should be of a technical/productive nature, i.e. you product things that you can show to us and we can show to others. Ideally, the analysis is done before GSOC start so that you can hit the ground running.
- Heiko (github: karlnapf, IRC: HeikoS)
- Viktor (github: vigsterkr, IRC: wiking)
- Lea (github: lgoetz, IRC: leagoetz)
Totally depends on what you are after.
Most important requirements:
- you are extremely motivated and ready to work independently
- some Shogun and Machine Learning basics
- Docker (optional)
We are looking for cool ideas of what is possible with the help of Shogun. Imagine you had 3 months and 5000 USD, what would you do?
We imagine the project roughly like this: you define an interesting question, go out to find data that contains the answer, use Shogun to extract the answer and finally document and visualise the whole process. How exactly this project might look like will depend on what kind of question you want to answer and the data involved - it could be anything from mapping bike accidents in cities, to researching the link between national parks legislation and biodiversity, to predicting election outcomes.
Projects are aimed to be self-contained. This means that they do not necessarily need to be integrated within the main part of the library. They do not even need to be C++ code. Consequently, you have more freedom when working as the sometimes hard-to-dodge quality checks for our framework do not need to be passed. At the end, we expect a Docker container with an installed (potentially modified) Shogun and code to reproduce everything that you did. In addition, we require a number of major in-depth blog posts about your project, and what problem you are solving.
As the project is of a much less technical nature than the others, communication with the mentor is slightly different: we expect you to work more independently (of course we help where we can), and there will be a larger and more frequent number of desiderables that will make up success/fail of the project.
Below is an idea for a project that will give you a flavour - but a flavour only. We expect you to get in touch with your own ideas (even if it’s preliminary - as long as you can demonstrate that you have thought them through). This is your opportunity to design your very own GSoC Data Project!
Are you a creative person, whose interests are beyond "implementing new algorithms", and who thrives in applying Machine Learning to real problems? Then this project is for you!
- Data Project - Elections
- Application Project - Interactive Webdemos
- Integrate Shogun with zepelin
- First step: a dataset or problem to work on, and an Ansatz
- Most important: a very detailed outline of the project goals and parts, and a detailed timeline
- Topics for 3 major blog posts about the project.
- Final step: Docker image to reproduce all results
- Map data examples
- Example of mapping bike accidents in London
- Example of spatial/temporal analysis of Twitter rumors and how it was done