Min/Max model constraints? #1217
-
|
Is there a way to set 'bounds/constraints' on the models? For example, I have a forecast that can never be below 0. Would there be a way internally to set such a constraint or should that be a post-processing step? Modifying after the fact would be simple enough but I worry about the feedback to the internal 'coefficients' in subsequent runs in the model doesn't 'know' that the previous negative forecast was 'out of bounds'. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
Hey there. There is
You shouldn't worry about that. If your model is well behaved (e.g. a sane learning rate for a linear model) predictions will make sense, and won't be negative when they shouldn't be. In particular, if you take tree models, it's guaranteed they won't produce negative outputs if they've never seen any negative label. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the clarification |
Beta Was this translation helpful? Give feedback.
Hey there. There is
PredClipperif you're too lazy to clip the predictions yourself.You shouldn't worry about that. If your model is well behaved (e.g. a sane learning rate for a linear model) predictions will make sense, and won't be negative when they shouldn't be. In particular, if you take tree models, it's guaranteed they won't produce negative outputs if they've never seen any negative label.