An example project showing how to create an H2O based machine learning application.
It consists of two different modules:
-
h2o-train : which depends on spark, and can fit a model on large datasets
-
h2o-predict : has no dependency on spark, and will be able to just run the model.
The model will be passed from the training module to the prediction module by serialization.
The preferred way is by using ModelSerializationSupport, but the model serialized by that class couldn’t be imported by MojoModel.load.
Therefore this branch holds an alternative implementation which seems to work fine.