I have a script in R that is frequently called during the day (by other scripts). I call R in a terminal using
Rscript code.R argument1 argument2
I notice it takes a lot of time to load packages and model and set up R.
Loading the model.rds in R is slow. This will affect the performance of my APIs. Ideally, I want to load the model once and then cache it inside the R environment.
Would it be possible to load up trained models (.rds)? How?
Thanks in Advance.
Hi,
I'm not sure whether I got your question rigth. Can you please specify the model.rds file? Does it including your trained model or are you currently building your model new every time you call code.R?
Or are you having several models? In that case you can save()/load() a list of objects by saving them as .RData. In some cases you might even want to export R's entire enviroment (without packages) with `save.image()`.
Further, if loading the packages slows you down, you can also skip loading a complete package and just specify the corresponding function (package::function()). This might help if you load a large package but use just a fraction of it.
Best.
I would suggest writing a top level wrapper script/function which loads everything if some global counter=1, then, it calls your actual script (periodically, or user input initiated) multiple times inside.
another option is to use an interactive shiny UI where you first load everything and then use buttons to re-run your specific test function/script, .e.g. I am running one here, which is loading a machine learning model (random forest ) in the beginning, and predicting/scoring real time on new test csv files uploaded.
https://virama.shinyapps.io/health_app3/
Hi,
Jeroen Staab and Vikas Ramachandra.
I am building a recommendation engine. So I only save my model in model.rds file from train.R script.
After that i load my model in test.R environment.Getting the prediction user by user.I have almost 3 GB model . So it takes 45 every time for loading and for prediction it takes only 0.90 seconds.
In my test.R environment
I call R in a terminal using
Rscript test.R userId numberofrecommendations
so In test.R i have load model and it takes 46 seconds almost for getting one user output.
So is there any other options to load my model one time initialise as service or make my model as daemon type. So i will get user based prediction of my engine very fast.
Thanks in Advance.
Hi, in this case I would opt for Vikas' answer. Both his approaches include a running R session and may be comparred to a deamon service. Keeping the session alive is the only way to keep your model in cache.
Best
- Badges
- Science topic