Hi Egor,

River has implemented AdaBoost. I think it depends on the data and application. If you have "enough" data and don't expect the relationship between input and output data to change much or change quickly, then a slightly stale batch-trained XGBoost algorithm could be an acceptable solution. In cases where you don't have much data and want a model to learn as you go, or where the relationship between input and output data to change, continual learning might be more appropriate.