How to handle NaNs in neural network training data?

I have NaNs in my neural network training data, and I was wondering if there was any way to deal with them when training? As of now, the network refuses to train. I found the following old links that say that there is not a whole lot to do, is that still true?

http://www.mathworks.com/matlabcentral/answers/95804-how-can-i-specify-missing-data-within-the-neural-network-toolbox http://www.mathworks.com/matlabcentral/answers/92319-how-are-nan-values-in-the-input-data-for-a-neural-network-taken-into-account-while-training-the-netw

Thanks in advance for any help!

Answer 1

[Greg Heath]

Do input and target NaNs match up?

Are the NaNs isolated or do they occur in batches?

If they are isolated you can look at the statistics of the diff function between NaNs

I am not experienced with this. However, I would try to interpolate the missing values. I would think this is better than replacing with average values.

First, however, why don't you experiment with a small data set to see how the processing function "fixunknowns' works

Hope this helps.

Thank you for formally accepting my answer

Greg

Answer 2

[Nipun]

I will try that and report back, thank you!