Things that might deserve listing in what's new and appear absent (from maint_tools/whats_missing.sh 0.20.3 0.21):

• ENH Convert y in GradientBoosting to float64 instead of float32 ([MRG+1] Convert y in GradientBoosting to float64 instead of float32 #13524)
• FIX _estimate_mi discrete_features str and value check (Fix _estimate_mi discrete_features str and value check #13497)
• FIX GBDT init parameter when it's a pipeline ([MRG] Fix GBDT init parameter when it's a pipeline #13472)
• Fix bug in AffinityPropagation ([MRG+1] Fix bug in AffinityPropagation #13334)
• ENH: process DataFrames in OneHot/OrdinalEncoder without converting to array ENH: support DataFrames in OneHot/OrdinalEncoder without converting to array #12147 ([MRG] ENH: process DataFrames in OneHot/OrdinalEncoder without converting to array #12147 #13253)
• FIX Raise warning in scikit-learn/sklearn/linear_model/cd_fast.pyx for cases when the main loop exits without reaching the desired tolerance ([MRG+1] Raise warning in scikit-learn/sklearn/linear_model/cd_fast.pyx for cases when the main loop exits without reaching the desired tolerance #11754)
• FIX Fixes tree and forest classification for non-numeric multi-target ([MRG] Fixes tree and forest classification for non-numeric multi-target #11458)

FIX GBDT init parameter when it's a pipeline (#13472)

That one doesn't need an entry since it's only an additional fix to
#12983

ENH Convert y in GradientBoosting to float64 instead of float32 (#13524)

Maybe something like:

- |Fix| :class:`sklearn.ensemble.GradientBoostingClassifier` and
  :class:`sklearn.ensemble.GradientBoostingRegressor` now keep the
  input ``y`` as ``float64`` to avoid it being copied internally by trees.
  :pr:`13524` by `Adrin Jalali`_.

Thanks @NicolasHug

Thanks for looking over this @NicolasHug