Simplify a CCA/BIBFA/GFA model
Prunes out unnecessary components and determines for each of the remaining components whether it is shared or not. In other words, the function reveals the component allocation into shared and view-specific ones.
A list of model parameters as returned by
The proportion of relative variance explained that components need to exceed to be detected as active.
This function can be used to prune out unnecessary components and
to recognize which of the components are shared. This can be useful
for interpretative purposes, but it is typically not necessary to
apply this function prior to making predictions (with
GFApred or otherwise). The inactive components will
anyway automatically cancel out for the predictive formulas.
The code works well for low-dimensional data, but for complex high-dimensional
data sources one should check whether the trimming is reasonable;
in such cases it is difficult to make clear decisions on component activity.
A list of parameter values as returned by
The list also includes two extra elements:
A boolean variable indicating that the model has been trimmed with this function.
A binary matrix indicating for each component (column) in which views (row) it is active.
Seppo Virtanen and Arto Klami
1 2 3 4 5 6 7
# # Assume we have a variable model which has been learned with # GFAexperiment() or GFA(). # Then the following line would trim it: # # trimmed <- GFAtrim(model) #
Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.