Limit of prediction

I am predicting continuous values (regression problem) and I have limits for the true values
but sometimes algorithm predict values exceed those limits
is there a way to enforce algorithm, (whatever is) not exceeding a threshold, but instead providing the limit value?

No, not in my experience.

1 Like

I’ve never tried to do something like that either, but you get to define the output layer “activation function”. If the model output is a continuous real number and you wanted to make sure that it’s never greater than 42, you could try using a function like this as the output activation:

def outputWithLimit(z):
    return np.minimum(z, 42.)

Of course you would then need to keep in mind that the derivative of that function is needed as part of your gradient calculations for backprop.

Seems worth a try anyway. Let us know if this type of idea helps for whatever it is you are trying to build.

1 Like

it seems good for NN
sure I will try it