netmap.model.nbautoencoder.NegativeBinomialLoss

class netmap.model.nbautoencoder.NegativeBinomialLoss(*args: Any, **kwargs: Any)

Bases: Module

__init__(scale_factor=1.0, eps=1e-10)

Negative Binomial loss module.

Parameters:

• scale_factor (float) – Scale factor applied to predictions. Defaults to 1.0.

• eps (float) – Small value for numerical stability. Defaults to 1e-10.

forward(y_true, y_pred, theta)

Compute the Negative Binomial negative log-likelihood loss.

Parameters:

• y_true (torch.Tensor) – Ground truth counts (non-negative integers).

• y_pred (torch.Tensor) – Predicted mean values (mu).

• theta (torch.Tensor) – Dispersion parameter.

Returns:

Mean negative log-likelihood of the NB distribution.

Return type:

torch.Tensor