Fast and scalable `agtboost`
Past due by almost 3 years
20% complete
agtboost should in principle be as fast and scalable as e.g. xgboost.
First steps towards this is
- Fix memory consumption problems
- Avoid costly re-building vectors of size-n iteratively and recursively
- Consider the design of classes
- Lean code: Remove un-used/experimental features
- Certain parts are embarrassingly parallell -- Employ OpenMP
agtboost should in principle be as fast and scalable as e.g. xgboost.
First steps towards this is
- Fix memory consumption problems
- Avoid costly re-building vectors of size-n iteratively and recursively
- Consider the design of classes
- Lean code: Remove un-used/experimental features
- Certain parts are embarrassingly parallell -- Employ OpenMP