explain need for normalizing in reweight docstring

hep_ml/reweight.py

@@ -16,6 +16,12 @@
 Algorithms are implemented as estimators, fitting and reweighting stages are split.
 Fitted reweighter can be applied many times to different data, pickled and so on.
 
+Remark: The normalization constant in the reweighters is not fixed. This is to ensure that the
+output of `reweighter.predict_weights` is deterministic; for example, if you predict weights for
+a large sample all at once or predict weights separately for each event and then concantenate the
+predictions, the result will be the same---the results would be different were the weights automatically
+normalized to the number of events. If normalization plays a significant role in your application,
+you should normalize the weights yourself.
 
 Folding over reweighter is also availabel. This provides an easy way to run k-Folding cross-validation.
 Also it is a nice way to combine weights predictions of trained reweighters.