- Store the # of samples seen at the leaf & split the leaf with gini impurity thresholds
Concerns: How do determine the splitting when only access to current batch?
- Store the overall accuracy of trees across different training batches, use it to find the weakest tree to replace
Concerns: bias towards the current sample batch