How do I represent an empty data in a matrix?

I need to train a neural network with 10 inputs (for example). I want to predict the time what will take build a steel material, having datas from the past. So, for predict the time, I need datas from the material (length, width, height, density, etc) but not always have the same datas, I mean, sometimes don't have the density, for example (because someone forgot to write it down) but I have the time it took. So, in that case, when I have a empty data, I complete with a 0 or how I represent an empty field in a matrix?

[Stephen23]

NaN

Answer 1

[BERGHOUT Tarek]

if that data is very important for you , i propose this solution for you:

• before you train your model , take only the raws that contain all of the features and build a new dataset with them.
• use this data to train an auto-encoder based neural nets.
• for the rest of simples put '0' in each loosed parameter .
• then map this data throw the trained auto encoders , and this final will predict the missing valuses according to your training set (training set that has all the features in samples )
• then collect your new dataset from the the initial and the regenerated one and use it to train your model.
• and don't forget to accept the answer if it is helpful
• and also don't forget to enjoy with this experience .
• note: do not use the target during trainig of the auto-encoder becaused its an unsupervised learning.