Appends columns from a metadata matrix Z to the H factor before
passing downstream. The next layer's W learns to factor out the
conditioning variables.
Usage
factor_condition(input, Z)
Arguments
- input
An fn_node (typically an nmf_layer output).
- Z
Conditioning matrix (n x p) or (p x n). Will be oriented
to match H dimensions.
Value
An fn_node of type "condition".
Examples
data(aml)
inp <- factor_input(aml)
layer1 <- nmf_layer(inp, k = 5)
# Condition on batch metadata (2 batches)
Z <- matrix(rep(c(1, 0, 0, 1), c(60, 75, 60, 75)), nrow = 135, ncol = 2)
conditioned <- factor_condition(layer1, Z)