netmap.model.zinbautoencoder¶

Zero-Inflated Negative Binomial (ZINB) autoencoder for scRNA-seq count data.

This module implements a ZINB autoencoder with three decoder heads for the mean (mu), dispersion (theta), and zero-inflation probability (pi) parameters. Forward-mode flags (forward_mu_only, forward_theta_only, latent_only, forward_pi_only) redirect ZINBAutoencoder.forward() to return a single tensor required by Captum attribution methods.

Functions

`get_mus`	Return the mean predicted mean expression (mu) across all cells.
`get_mus_grouping`	Return the mean predicted mu per observation group.
`get_thetas`	Return the mean predicted dispersion (theta) across all cells.
`get_thetas_grouping`	Return the mean predicted theta per observation group.

Classes

`ZINBAutoencoder`	Autoencoder with three decoder heads for ZINB-distributed scRNA-seq data.
`ZINBLoss`