I know you can generate Java POJOs from H2O models built in Python or R, but is there a way to build C# dlls or something similar that can run in a .NET environment? I have seen references to this but nothing concrete.
2 Answers
The R, Python and JavaScript/CoffeeScript (i.e. Flow) clients all just make REST API calls to the H2O server. If your production workflow allows running the H2O java server in the background, you could load models and call predict on them, from your C# application, by making those same REST API calls.
It is well documented but it might be easier to do the desired steps in Flow, while watching the network calls being made.
Whether this is better than your own idea to export the model parameters from H2O and import them into a C# native library is open to debate, of course. Running the java server and using API calls is more likely to come out ahead if you are batching predictions.
Related