Hi all,
I was recently wondering about how machine learning systems are developed in real world, especially when there could be new data streams and new features in the future.
Imagine, there is a customer marketing model, which uses data from 20 separate sources and integrates them together and learns a classifier/predictor. Now six months down the line, there is a possibility of adding another set of features which were previously not explored in the model.
How does the integration take place ? In a batch setting, is the model recreated from scratch, assuming missing values for data not available ? or is there a way to update the model without rerunning the entire system ?
[link][3 comments]