A GBDT can have data added or removed in place, without full retraining, by updating only affected tree nodes and stored split statistics.
Similarly, W2C represents the testing instances that are wrongly predicted during retraining but are correctly predicted after decremental learning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Online Gradient Boosting Decision Tree: In-Place Updates for Efficient Adding/Deleting Data
A GBDT can have data added or removed in place, without full retraining, by updating only affected tree nodes and stored split statistics.