2

When consulting, I often do my exploratory analysis and prototyping in R, and deliver results on the initial dataset to the client. The client wants to use the trained classifier in a production setting. For example, assume that the client wants to run 1000 records per day through the classifier. How should I deliver the trained classifier?

For example:

  1. Send paper describing the results and R code that will exactly replicate my work and hope that the client will be able to make a production server and implement it properly?

  2. Abandon R after the proof of concept, and write up the classifier in a different programming language (e.g. python) that is more readily deployable by the client?

  3. Try to set up a webserver that will run an instance of R?

  4. Accept data from the client and run it upon request?

greg
  • 21
  • 1
  • Try R-Shiny? http://rstudio.github.com/shiny/tutorial/ – curious_cat Apr 03 '13 at 22:05
  • Is real time response needed or can all 1000 records be batched together? – curious_cat Apr 03 '13 at 22:07
  • Would it be possible to deliver a PMML format ? – Giorgio Spedicato Apr 03 '13 at 22:31
  • R-Shiny is a good suggestion -- I browsed through their page earlier today and will probably sign up for the beta test (otherwise, it looks like it falls into the "run your own webserver" approach, though with a better interface than most) – greg Apr 04 '13 at 00:00
  • In my case a real-time response is desirable but it's not strictly necessary. I'm definitely not above accepting data over email and running it myself, but somehow that doesn't seem like the "right" answer. – greg Apr 04 '13 at 00:05
  • Why not ask the client which one of {1,2,3,4}, or perhaps another option, they prefer? – arielf Apr 09 '13 at 22:22

0 Answers0