Building a statistical significance testing web service powered by R | Hacker News Reader

Building a statistical significance testing web service powered by R

garysieling.com

20 pointsgarysieling10 comments

I've heard good things about the package Rserve. It's a good way to combine a lot of server side R requests with your online app of choice. http://www.rforge.net/Rserve/doc.html

There's also an R buildpack for Heroku. There's a sample app written using Sinatra (Ruby) included.

https://github.com/virtualstaticvoid/heroku-buildpack-r

Would it be possible to code everything in something like Django to handle the HTTP, view parts, and farm out the calculations to R?

How would you design that?

Of course, but at that point you likely wouldn't be using any R-leaning webservers; rather, you'd start with any old web server, and then just make shell calls to R. Or, you _could_ do the "R as a service" type thing, and have your normal [Django] stack make a webservice call to the R-webservice.

Yes, you can use rpy/rpy2 in your python code to delegate stats to R

I made one in pylons. It kept blowing up in the memory. It was a bitch to get up and running and then debug. And then, well, I wasn't sure what I couldn't do with highcharts and python libraries instead.

That's the general direction I am heading, although I haven't picked a front-end yet. I'm thinking of returning the more complex results as JSON, and using ExtJS for grids, charts, etc.

Keep in mind that R is an in-memory language and that each object you'll be creating will be stored in RAM. This could be tricky if you have huge datasets

This is really neat! You should post more examples, like if I had a range of data points could I post that data set and have R output a best fit line?

Thanks, glad you liked it, I have more examples coming.

There is a library you can install that looks like a web-based charting frontend for R, although I have not had a chance to try it-

http://www.stat.ucla.edu/~jeroen/ggplot2/

Reply on news.ycombinator.com