NegativeLogLikelihoodLoss

class mlquantify.losses.NegativeLogLikelihoodLoss(reduction='mean')

Negative log-likelihood loss for mixture likelihoods.

Computes \(-\log p(x)\) element-wise and then reduces the resulting values by mean or sum.

Parameters:

reduction{‘mean’, ‘sum’}, default=’mean’

How to reduce the per-sample log-likelihood values.

Attributes:

reductionstr

The configured reduction mode.

See also

MixtureNegativeLogLikelihoodLoss

Builds the mixture likelihood from per-class likelihoods.

RegularizedMixtureNLLLoss

Adds simplex-smoothness penalties.

get_loss

Factory that builds this loss by name.

Examples

>>> from mlquantify.losses import get_loss
>>> loss = get_loss("nll")
>>> import numpy as np
>>> loss(np.array([0.8, 0.6, 0.9]))  
0.2797765635793423

__call__(likelihood)

Compute the negative log-likelihood.

Parameters:

likelihoodarray-like of shape (n_samples,)

Per-sample likelihood values in the range (0, 1].

Returns:

lossfloat

Reduced negative log-likelihood.

Examples

>>> from mlquantify.losses import get_loss
>>> import numpy as np
>>> loss = get_loss("nll")
>>> round(loss(np.array([0.5, 0.5])), 4)
0.6931