I observed that logp values are positive for some cases, when calculating the WAIC. In my understanding the logp when computing WAIC should be the log predictive probability from the predictive density. But is not it a problem having probabilities greater than 1 ?

You can have logp values that are positive - as the pdf will return value that > 1. Dont confuse that with probabilities greater than 1 - probabilities is a measure of area within the small interval in the continuous cases.

Can logp be positive when computing WAIC

Questions

Nadheesh May 17, 2018, 6:38am 3

Ok… I understand it now. Thanks for the clarifications.

Topic		Replies	Views
Calculating model comparison using log pointwise predictive density Questions	15	3597	April 5, 2018
Using WAIC to compute Log Predictive Accuracy Questions	3	751	May 11, 2018
Unexpected Results when using WAIC to Compute Log Predictive Accuracy Questions	3	688	May 13, 2018
Evaluate logposterior at sample points Questions	9	4411	August 16, 2017
Calculating WAIC for models with multiple likelihood functions Questions	20	4626	October 26, 2020

Can logp be positive when computing WAIC

Related topics