Add subset correction for target manifold dimensionality in the feature selection algorithms
The number of selected variables should not be less than the user-specified target manifold dimensionality, even if minimizing the cost suggests so. Perhaps we can just print a message that the resulting subset is not the one that minimizes the cost and print a recommendation for the user to consider lowering the target manifold dimensionality.
This affects:
analysis.manifold_informed_feature_selection()
analysis.manifold_informed_backward_elimination()
Edited by Kamila Zdybal