Rory Mitchell
Rory Mitchell
Focusing on the `RegTree::CalculateContributions` part for now. Why does the lambda regularisation term appear in this formula? It's not involved in prediction at all. When you write $\frac{|L|}{N} \frac{\sum_{X \in...
When you make the prediction $f(X_i)$ the lambda term is already baked into this return value. You don't have to include it.
I think I see what you mean. I agree that `base_weights` should be used to exactly implement Saabas' method according to the blog post. If our approximate contributions method exists...
Benchmarks look good: | dataset | master | integer | |------------|-------------|-------------| | airline | 58.75668815 | 59.16649409 | | bosch | 12.61275904 | 10.81264534 | | covtype | 14.17920381 |...
This does not fix anything yet, I need to do this second part which affects the calculations in split evaluation. This spark test failure looks unrelated to this PR.
I think I need to reduce over the distributed workers in the histogram rounding bound calculation.
Allreduce is now happening on quantised gradients. We would get the wrong results if each worker provided integer gradients quantised in a different scale.
@trivialfis can I please get another review?
I don't think any current active maintainers are big R users so we welcome input. Could we just build a new interface behind a different namespace until it's ready? I...
Thanks for the reminder. Maybe I should test on Ampere to check that I haven't reintroduced that issue. I think the number of blocks launched should be even smaller in...