There’s a possibility I may be looking to implement some sort of interface (web or otherwise) for various tasks related to statistical models/simulations that I will build. I think, if I did do this, it would be with the next year sometime after handing in my dissertation and before getting a job. I’ve been thinking a bit about what such a project would really need, what I would like, and how to put this together to make it fairly usable to both the mathy maintainer (me) and a potentially fairly wide array of users. Here are my thoughts an notes so far. They probably don’t make too much sense as I haven’t done anything much like this before.
Tasks:
- Data Collection (polling): There is some model that takes a large number of input parameters. Some parameters are “known” by modellers while others are estimated by polling experts. Polls are conducted in a funny way that ranks voters according to some weighting algorithm and computes weighted estimates of parameters based on the weighting scheme. I would like voters to be able to draw curves when answering questions, so some sort of graphical interface would be great … I’ve seen a few java applets that seem to do this reasonably well.
- Data Management and Data Entry (modeller): Given a set of polling data, the modeller computes the set of parameter estimates that they need. They also input the other parameters they need for the model. They hit *save* and name the input-set something and voila, they are ready to run simulations. The modeller can have multiple input sets to test different ranges of parameters.
- Running Experiments/Simulations, Output Management, Post-Processing etc.: Simulations are run and output stored for post-processing and analysis.
Polling through a web-interface seems to make sense. After that, the other two tasks can occur offline. For simple, restricted types of simulations and simulation output I could imagine a web-interface but that would be at very late stage of this hypothetical project.
Ideally, in the input collection stage I would like to offer the ability to produce weighted surveys from any selectable subset of users. This means (I think) finding a nice web-interface to a database. I will likely collect and store these through some sort of xml file. See
- http://www.omegahat.org/StatDataML/
- http://www.w3.org/TR/SRML/
- http://www.ilog.com/products/jrules/documentation/jrules67/rstestsim/rs_rsm_refxmldefine4.html
Levels of users or usage:
- developer (code management)
- admin (full data managment)
- modeller (run models, extract data, manage output data only, create subsets of input data)
- data provider (I’m thinking of data coming from polling experts here)
Only the developer and admin class should have destructive access. It would be nice to somehow allow modellers and providers to maintain accounts and manage their profiles through a reasonable interface. This might only be practical for data providers.
This is probably just a “thoughts to myself” page … but any advice, especially on the polling interface, is welcome.
