What is 'deviance' and why is it used to evaluate GLMs?
hardAnswer
- Deviance = -2 * (log-likelihood of your model - log-likelihood of the saturated model).
- Analog of squared error for non-Gaussian likelihoods: Poisson deviance for count data, Tweedie deviance for insurance-style claims, binomial deviance for logistic regression.
- It's the natural loss for the corresponding GLM and gives coherent training + evaluation.
- Modern GBM libraries expose Poisson / Tweedie / gamma deviance losses directly — pick the one matching your target distribution.
Check yourself — multiple choice
- Deviance is only for classification
- -2·(log-likelihood − log-likelihood of saturated model) — natural loss per exponential-family distribution
- Deviance equals
- Deviance is always MSE
Deviance = analog of squared error for the GLM's exponential-family likelihood.
#metrics-regression#linear-models#theory
Practise Supervised Learning
215 interview questions in this topic.
Related questions
- What's the Bayesian interpretation of Ridge regression?
- What is a Generalized Linear Model and when is it the right tool?
- L1 vs L2 regularization — what's the difference?
- Why is logistic regression called a linear model if it uses a sigmoid?
- Generative vs discriminative classifier — what's the difference?
- What is the curse of dimensionality?