decoder_small: Network small
In scAlign: An alignment and integration method for single cell genomics

Defines decoder network architecture for scAlign.

1
2
3

decoder_small(inputs, complexity, final_dim, emb_size = 32,
  l2_weight = 1e-08, dropout = TRUE, dropout_rate = 0.3,
  is_training = TRUE, batch_norm = TRUE, shared_ae = FALSE)

`inputs`	Mini-batch placeholder
`complexity`	Determines the depth and width of an automatically created network
`final_dim`	Number of features in high dimensional data
`emb_size`	Number of hidden nodes in final (embedding) hidden layer
`l2_weight`	Weight on l2_regularizer
`dropout_rate`	Probability for dropout.
`is_training`	Determines if dropout and batch norm should be include in pass through network
`batch_norm`	Determines if batch normalization layers should be included