Tianyi Zhou, Tianqi Chen, Luheng He {tianyizh,tqchen,luheng}@uw.edu
- Tianyi leads the project report, presentation, and develop parts of featureview module.
- Tianqi leads machine learning backend server support, and develops the pathview module.
- Luheng leads the implementation of interaction interface, and communication with backend.
- All the team members are involved in discussion, and development of ideas:)
http://cse512-14w.github.io/fp-tianyizh-tqchen-luheng/
- Start with toy dataset from UCI machine learning repository for prototyping our project.
- Use static data exported from the machine learning tool for front-end development, such as visualization of tree and path.
- We added another larger dataset from Kaggle in the health domain to see if this tool can help difficult machine learning problems.
- For the next step, we implemented frontend-backend communication to enable user interaction and instant feedback of the machine learning algorithm.
- From exploration with the current tool, we decided to add more powerful features such as feature selection (both groups and single features) and generalized tree modification. Both frontend and backend need to make big changes to support these new interations.
The demo can run in linux or mac machines, first make make sure system requirements are met
- g++: we need g++ to compile backend machine learning algorithms
- screen: used for creating backend request handling server
- python: used for backend script
Make sure the machine is connected to the internet, run: ./install.sh
If the system requirements are met the script will do everything, and you can open browser in localhost:8000 to use the demo.
If you exit the cgi server, e.g. ctrl+c, use ./startserver.sh to start it. However, there is a chance that you need to wait less a a minute to run ./startserver.sh after you exit it. This is due to the fact that when local tcp server exits and, the local port used by mlserver need sometime to be released.