ndarray.has_nan(NDArray) and its convenient form NDArray.has_nan(), which return True if there is nan in the data else return False. This is very helpful in debugging.

Optimally has an option to output the per location prediction ndarray.has_nan(x, gather=True), when gather=False, it should output a binary NDArray with the same shape as x and each location predicts the corresponding data is nan or not.