I am trying to find a way to use R to fit neural network models, and then use it to score in SAS. To give some context, we currently have a modelling pipeline (fitting and scoring) created entirely in SAS, and the problem is the model fitting is taking too long. Instead of paying a fortune to get more SAS CPU cores and the license to use them, the plan is to run some of the models using R to spread the processing load to the R infrastructure.
The main constraint is that the scoring module is written in SAS, which essentially scores the data by including all the text scoring code in a single giant data step. So the output from R needs to be compatible with this scoring module.
I am not too familiar with R, but it doesn't seem to generate a text scoring code. If this is in fact true, then the alterative would be to use the weights/bias/activation functions to manually score the data, which isn't very feasible.
My question is whether there are more feasible approaches?
Thanks
