I tried to post the question below on Stack Overflow, then on Cross Validated, then on Server Fault but my question has been stated as off-topic each time. I'm a bit fed up because I have posted far more answers than questions on Stack Overflow and the day I need some help: no way. I always asked how could I make this question not off-topic or where would it be better to post then, but no answer. Just a downvote or on-hold response. I know it is not a programming issue but there are many very, very similar questions on Stack Overflow so I am really surprised. It seems a bit more picky when it relates to questions on R.
R web-service for real-time prediction with online learning
What free framework or R packages should I use to deploy R as a web service handling many models for real-time prediction (with online learning) ?
It should be able to address those key points:
- allow parallel processing/requests while handling a memory/cpu policy preventing reaching memory/cpu limits.
- update many models in parallel safely (ie. prevent concurrent updates). able to quickly load/unload many models on a single machine because not all models (ie. >1000) could fit in memory.
- handle large streams with low footprint (out-of-core).
- able to use VowpalWabbit and StreamMOA R packages.
- web service could be Http or Thrift based.
Thanks a lot for any help!
FYI, the different versions of my question:
- 1st try on SO: https://stackoverflow.com/questions/26488144/r-for-real-time-prediction-and-online-learning-on-many-models
- CV: https://stats.stackexchange.com/questions/122748/r-web-service-for-real-time-prediction-with-online-learning
- 2nd try on SO: https://stackoverflow.com/questions/26775103/r-web-service-for-real-time-prediction-with-online-learning
- Server Fault: https://serverfault.com/questions/642516/r-web-service-for-real-time-prediction-with-online-learning