Using model simulator to score a model

Chemical_eng
New Altair Community Member
Hello,
I am using model simulator to score a model . Sometimes when I want to do this one of the input parameters is marked in red, I guess because it is outside the training range ? Can you tell me why ? What is the interpretation or consequences of doing this ?
I am using model simulator to score a model . Sometimes when I want to do this one of the input parameters is marked in red, I guess because it is outside the training range ? Can you tell me why ? What is the interpretation or consequences of doing this ?
Tagged:
0
Best Answer
-
Hi!
Exactly, you get the red text when entering a value outside of the training range.
Different machine learning methods cope more or less well with exceeding the training range. For example, tree and nearest neighbor methods are often not very good at extrapolation.
Linear regression is often OK. Non-linear ones, neural networks etc. can have expontential effects that are again unsafe when extrapolating.
You can test this. Instead of randomly selecting a test set, you can filter for high values and use that as the test set. Then you can validate the extrapolation capabilities by testing your model on these examples. (This is not a generally reasonable way for validation. But it can help you with testing this particular property of the models.)
Regards,
Balázs0
Answers
-
Hi!
Exactly, you get the red text when entering a value outside of the training range.
Different machine learning methods cope more or less well with exceeding the training range. For example, tree and nearest neighbor methods are often not very good at extrapolation.
Linear regression is often OK. Non-linear ones, neural networks etc. can have expontential effects that are again unsafe when extrapolating.
You can test this. Instead of randomly selecting a test set, you can filter for high values and use that as the test set. Then you can validate the extrapolation capabilities by testing your model on these examples. (This is not a generally reasonable way for validation. But it can help you with testing this particular property of the models.)
Regards,
Balázs0