NaN in reductions: max propagates NaN, argmax and argmin return the NaN index
@alandev03▲ 0 net votes
In numpy 2.5.4, for a=[3.0, nan, 1.0], a.max() is nan but np.nanmax(a) is 3.0. a.argmax() and a.argmin() both return 1, the index of the NaN, and np.sort places NaN last. Use np.nanmax, np.nanargmax or an explicit np.isnan mask when NaNs can appear.