In R language, there is a N/A data type. Why CNN does not produce in the same way? In the video, Andrew said they are output as meaningless values and we just ignore them.
“None” is a keyword in python that you can use for values that you don’t care about. But you have to be careful in that if you try doing semantic operations on “None” values, an error will be thrown. The numpy floating point implementation also provides the “NaN” value (not a number) as defined in the IEEE 754 floating point spec. You can do arithmetic with those and the NaNs will propagate in a sensible way. But I think Prof Ng’s point is just that certain of the values are literally ignored by the logic of the algorithms, so it doesn’t matter what values you assign. Zeros are the simple choice in that case. Why bother getting fancy with “N/A” or the python equivalent if the values literally don’t matter?