Add a builtin ML implementation

[ggeorgakoudis]

• Add implementation of decision tree, random forest
• Make OpenCV an optional dependency (ENABLE_OPENCV)
• Add env var (APOLLO_TUNING_MODEL) to select DecisionTree or RandomForest models
• Add python scripts to visualize with models with dot
• Update Kokkos connector

[cdwdirect]

Yes. I agree about implementing an internal DT.

On Nov 2, 2021, at 8:09 PM, Giorgis Georgakoudis @.***> wrote:

 @ggeorgakoudis commented on this pull request.

In src/Apollo.cpp:

           reg->time_model =

ModelFactory::createRegressionTree(train_time_features, train_time_responses); +#else

• assert(false && "Retraining requires OpenCV"); I agree that handling is not ideal. The ifdef-conditional assert is intentional. We do not (yet) support regression in our built-in ML library. The assertion will abort execution if the user enables re-training through the env var but has not compiled with ENABLE_OPENCV. Could we accept as a stopgap until regression falls in?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.