NegativeBinomialDecoder

latent.modules.decoder.NegativeBinomialDecoder

Decoder with negative binomial reconstruction loss. Uses size factors to deal with count data.

__init__(self, x_dim, name='nb_decoder', loss_name='nb_loss', hidden_units=[128, 128], initializer='glorot_normal', dispersion='gene', **kwargs) special

Parameters:

Name	Type	Description	Default
x_dim	int	Integer indicating the number of dimensions in the input data.	required
name	str	String indicating the name of the model.	'nb_decoder'
hidden_units	Iterable[int]	Number of hidden units in DenseStack. If set to None the model skips the DenseStack and reduces to a linear decoder (Svensson 2020).	[128, 128]
loss_name	str	String indicating the name of the loss.	'nb_loss'
initializer	Union[str, Callable]	Initializer for the kernel weights matrix (see keras.initializers)	'glorot_normal'
dispersion	Union[Literal['gene', 'cell-gene', 'constant'], float]	One of the following: 'gene' - dispersion parameter of NB is constant per gene across cells 'cell-gene' - dispersion can differ for every gene in every cell 'constant' - dispersion is constant across all genes and cells float - numeric value of fixed dispersion parameter	'gene'
**kwargs		Other arguments passed on to DenseStack.	{}