Nullable one-hot encoding

Can I have a one-hot encoding like [none, 1, 0, 0, none]; i.e. I don't know if the sample class is 0 or 4, but I know its class 1, but not class 2 or 3

The loss function would then ignore where there's a none so we don't provide false-negatives.

https://stackoverflow.com/questions/67029738/nullable-one-hot-encoding April 10, 2021 at 08:24AM