The task is to use XGBoost regression model, built in R (RStudio), in example, in Excel or similar, so that other user just loads new input parameters (numbers) and gets a number result (from the earlier built XGBoost model), not dealing with R/modelling/XGBoost etc.
May be in Excel, Access, sql, or standalone. Any ideas?
Yes, I have a simple solution for this, that I was considering writing up for on here if there is sufficient interest.
Basically, I have invented a simple, concise matrix format for representing sums of trees, and simple
efficient algorithms for (1) extracting the necessary information from an xgboost dump, and (2) computing and applying the trees/forests to new data. [so far I have just scripted these in MATLAB, SCILAB, and Python, but they are easy to port and tiny (a matter of kilobytes in size)]. Is anyone else interested in this?
Yes, I am anyone else interested!
I am interested to be able to parse the rules extracted from an xgboost model to make a native matlab routine. Please share your experience!