11-17-2016 08:17 AM
As long as the models are built on the same underlying superset of data attributes, and they are all trying to predict the same label, the fact that they were built on different datasets doesn't matter. You can simply store the models in the repository (using the "Store" operator) and then apply them all to a new dataset (using "Apply Model") as long as it contains all the individual attributes required. Note that you'll need to make sure any data preprocessing that was done for the original modeling also is replicated (e.g., any feature engineering or data transformations). You could then use them as an ensemble model (e.g., using the "Vote" method) for future predictions.
If the models are not built on some unified underlying superset of attributes to predict the same label, then I am not sure exactly what you would mean by "merging" the models. They would at that point be totally separate models and there really wouldn't be anything to merge.