ANU Data Mining Group

This web service is based on a predictive modelling framework using sparse grids (PMSG). For complete idea about this project, please refer to Markus Hegland and Margaret Kahn's slides (pdf, 237KB) from the APAC03 Conference.

Procedure

1. Web client sends query to web server.
2. Web server generates request and sends it to HPC server.
3. HPC server processes request, downloads data and sends results back.
4. Web server produces web page and sends it to client

Characteristics and Challenges

• Parallel execution using batch queue:
  • Assembly of linear systems of equations.
  • Evaluation on large data sets.
• Different time constants:
  • Instantaneous response of web server.
  • Batch jobs on HPC nodes.
  • File transfers.
• Persistent ssh tunnel and server process for continuing service.
• New asynchronous event-driven programming skill to enable several users to access the remote server simultaneously without delay.

How to use

1. Choose data set and structure of model.
2. Get information about data (metadata) by clicking the "Show Data" button.
3. Fit different models to same data through adding different components.
4. Display: data, prediction model and data vs prediction.

Note: the remote server and ssh tunnel need to be started first before running this application. To avoid leaving a server running, we created a simple demo. It is not a real application and you cannot expect it outputs the correct results. It is just for giving you some ideas about how it works. If you do want to run it interactively, please contact us to get it setup beforehand and then follow the link here.