netcal.metrics.regression.NLL

class netcal.metrics.regression.NLL

Negative log likelihood (NLL) for probabilistic regression models. If a probabilistic forecaster outputs a probability density function (PDF) \(f_Y(y)\) targeting the ground-truth \(y\), the negative log likelihood is defined by

\[\text{NLL} = -\sum^N_{n=1} \log\big(f_Y(y)\big) ,\]

with \(N\) as the number of samples within the examined data set.

Note: the input field for the standard deviation might also be given as quadratic NumPy arrays of shape (n, d, d) with d dimensions. In this case, this method asserts covariance matrices as input for each sample and the NLL is calculated for multivariate distributions.

Methods

__init__()
measure(X, y, *[, kind, reduction])	Measures the negative log likelihood (NLL) for probabilistic regression models.