Model Ensemble in PMML

Once a solution used by just a few data scientists, model ensembles are now being used to solve more and more problems. That's due in part to the highly publicized Netflix prize and in part to the many tools that now make it easier for users to develop a solution containing multiple models.

In a model ensemble situation, every model is executed and the overall result or output is a combination of the partial results obtained from each model.

PMML is capable of representing not only model ensembles, but also composition, segmentation, and chaining. The same is true for ADAPA, which can consume PMML files containing multiple models.

Although you can easily find PMML examples for single models, it is not so easy to find PMML examples for multiple models. To remedy this situation, we have made available quite a few multiple model examples in the Zementis website. For the most part, they cover model ensemble and segmentation and include random forest and random survival forest models exported in PMML with the R PMML package.


All PMML examples can be downloaded together with a CSV file containing scored data. The models can then be directly uploaded in ADAPA and the data used for model verification and scoring.


Article is closed for comments.
Powered by Zendesk