How to build a stacking regressor in Knime?

The regressor may have a few base models, such as a linear regressor, a tree model…

Any existing workflow can be referred to?

I am not aware of a workflow example but in general i assume you first do predictions for the individual models then use column appender (or similar to combine the predictions and then use another linear learner with the predictions as features and the original target column for your prediction
br

@HY_Z H2O.ai Automl would support stacked models. You can use the KNIME node:

Or you could use the packages wrapped with a R or Python node in KNIME which gives you more flexibility concerning the settings:

1 Like

thanks for the info. Unluckily I dont have H2O.auto ML.

thanks for the idea. Will try.

Well you have if you install the KNIME H2O Machine Learning Integration – KNIME Hub with the H2O AutoML Learner (Regression) – KNIME Hub

https://docs.knime.com/latest/analytics_platform_installation_guide/index.html#_installing_extensions_and_integrations

1 Like

thanks. will check. does that mean we can use H2O extension node even when we dont have H2O model?

@HY_Z I am not 100% sure I understand what you are asking. YOu can use the H2O.ai implementation in KNIME and build models and use them. H2O also does have a commercial product the models of which would not be compatible to be used with the free version.

Maybe you check the examples and try to describe in more details what you want to do.

1 Like

@mlauber71: Thanks a lot. I download one example which uses H2O auto.ml and have a try just now. Amazing, my own data can be linked to the H2O.automl learner node and the regression model can be built.

will continue to try the stacking regression function.

Amazing, isn’t it :slight_smile: handling and analysing data and building models and so on is the whole point about knime.

If you want to learn more about knime and machine learning you might want to check these resources.

1 Like

@mlauber71 thanks for the explanation. The meta collection is with a lot of topics. will go through. thank you.