# Should I train a neural network with data with or without a constraint?

I want to train a Neural Network (NN) using a dataset. I want to use the NN model as a prediction function in one algorithm. However, in the algorithm, any data that does not meet a specific constraint (say some parameter $$\theta <10$$) would not be included.

So, my question is, while generating the training data, should I include all kinds of inputs irrespective of the constraint, or should I generate only those data which meet the constraint $$\theta <10$$?

Currently, I am training data with constraint ($$\theta <10$$), and I am getting an average error of around $$6\%$$. Ideally, I want it below $$3\%$$.

I am new to NN model training. Any kind of pointers would be helpful.

• In general you want to train a model with a sample of data that represent as good as possible the data that the model will predict on when deployed. if you use a threshold in your first algorithm, then you're doing right in using the same threshold for your nn training. Regarding the error you should provide more info, first of all, average error of what? Oct 12 at 12:25