From adf87b27c52fb386fa01e5c417cee7f682d5b68f Mon Sep 17 00:00:00 2001 From: david-cortes Date: Tue, 20 Aug 2024 20:52:50 +0200 Subject: [PATCH] [doc] Fix tutorial for advanced objectives (#10725) --- doc/tutorials/advanced_custom_obj.rst | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/doc/tutorials/advanced_custom_obj.rst b/doc/tutorials/advanced_custom_obj.rst index b78cdc292..5b81b4716 100644 --- a/doc/tutorials/advanced_custom_obj.rst +++ b/doc/tutorials/advanced_custom_obj.rst @@ -360,7 +360,7 @@ point, which means it will be a minimum rather than a maximum or saddle point). But note that this is still not usable for XGBoost, since the expected Hessian, just like the true Hessian, has shape ``[nrows, k, k]``, while -XGBoost requires something with shape ``[k, k]``. +XGBoost requires something with shape ``[nrows, k]``. One may use the diagonal of the expected Hessian for each row, but it's possible to do better: one can use instead an upper bound with diagonal