link: https://en.wikipedia.org/wiki/S%C3%B8rensen%E2%80%93Dice_coefficient range:
summary
-
it is the log of the likelihood function
-
If we take the negative of the log likelihood, we get the negative log-likelihood (NLL) loss
- note: since the likelihood function changes depending on the distribution, there is no single NLL formula
- e.g. if we are doing a binary classification problem, we use BCELoss
-
Why use log likelihood instead of the pure likelihood function?
- https://math.stackexchange.com/questions/892832/why-we-consider-log-likelihood-instead-of-likelihood-in-gaussian-distribution
- since ln(a*b) = ln(a) + ln(b), many equations simplify significantly
- so we use the log likelihood over the likelihood
pros
- Log likelihood can be used as a score to detect anomalies or outliers in the data.
- By calculating the log likelihood of each data point under a trained model, data points with significantly lower log likelihood values can be considered anomalous.