encoderModel_add_semisup_loss_data_driven: Add walker loss component

In quon-titative-biology/scAlign: An alignment and integration method for single cell genomics

Description

Creates the walker loss component using embedding representation of high-dimensional data. Specifies random walk procedure.

Usage

encoderModel_add_semisup_loss_data_driven(
  p_target,
  a,
  b,
  data_shape,
  walker_weight = 1,
  visit_weight = 1,
  mode = "source",
  comp = "exp",
  betas = NULL,
  diag = "zero",
  debug = TRUE
)

Arguments

p_target	Pairwise cell-cell similarity
a	Data embeddings
b	Data embeddings
data_shape	Dimension of data
walker_weight	Walker loss component weight
visit_weight	Visit loss component weight (unused)
mode	Specifies which walker loss is being computed
comp	How to compute round trip probabilities
betas	Value for fixed beta gaussian kernel
diag	Whether to include self similarity
debug	Sets the verbosity level

Value

Tensorflow loss op

quon-titative-biology/scAlign documentation built on Nov. 17, 2021, 9:57 a.m.