2

I have a small RNN with a softmax output, which succesfully classifies sequences within a known set of n classes. The model is only trained with known classes.

Now I have the problem that there might be sequences which do not belong to any of said classes.

Of course, my model will try to lump in those sequences into one of the given n categories.

How to deal with these sequences that do not fit into any of the given/trained categories?

nilleeee
  • 53
  • 7
  • Have you tried to add *other* class? – Green Falcon Dec 14 '18 at 14:40
  • @Media I'm not sure what you mean. Add another class to the output? Add the class to the training data? Later is not an option. – nilleeee Dec 14 '18 at 14:45
  • I guess you didn't understand. In situations that there are other things in the input signal that you don't know them or you don't want to, one common thing is to gather real data distribution of those things that you don't want to classify and add them to your training data, and yes. you have to add another entry to your output for those inputs that you don't know them. – Green Falcon Dec 14 '18 at 14:48
  • Would you happen to have a keyword that describes this practice, so I know what to look for when I want to further dive into this? – nilleeee Dec 14 '18 at 14:53
  • [unkown](https://datascience.stackexchange.com/q/37706/28175). – Green Falcon Dec 14 '18 at 14:55

0 Answers0