1
$\begingroup$

In the R documentation for the cv.glmnet function, type.measure parameter is described as:

loss to use for cross-validation. Currently five options, not all available for all models. The default is type.measure="deviance", which uses squared-error for gaussian models (a.k.a type.measure="mse" there), deviance for logistic and poisson regression,...

Could someone help me with the exact definition of the term gaussian models? Does it refer to all regression models?

Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ errors are Gaussian $\endgroup$
    – Aksakal
    Commented Jun 11, 2020 at 0:35
  • $\begingroup$ pardon me, but is there a more layman explanation for what are Gaussian errors and what is non-Gaussian errors? I find it hard to grasp the meaning of the term... $\endgroup$
    – machine-building
    Commented Jun 11, 2020 at 0:51
  • $\begingroup$ Are you familiar with normal distribution ? $\endgroup$
    – Aksakal
    Commented Jun 11, 2020 at 0:54

1 Answer 1

Reset to default
2
$\begingroup$

Gaussian models are models that use the Gaussian distribution (also known as the normal distribution). The reference usually refers to an assumption that the "error terms" in the model are assumed to be normally distributed. Maximum likelihood estimation for this distribution can be shown to be equivalent to minimising the Mean-Squared-Error (MSE).

Improve this answer
$\endgroup$
2
  • $\begingroup$ But when would the gaussian assumption for 'error terms' be reasonable? For example can we assume all linear regression models have gaussian error terms? Thank you! $\endgroup$
    – machine-building
    Commented Jun 11, 2020 at 1:09
  • 1
    $\begingroup$ Theoretically yes, you can, If you assume that the errors are Gaussian, roughly speaking it's mean that errors are predictable. you may check prof Anderw Ng lecture notes page 11. (probabilistic interpretation).see.stanford.edu/materials/aimlcs229/cs229-notes1.pdf $\endgroup$
    – 4.Pi.n
    Commented Jun 11, 2020 at 1:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.