Is your feature request related to a problem? Please describe.
Not really a problem but a nice feature to have.
Describe the solution you'd like
Add the null-space correction, i.e., project out the null-space of the covariance matrix of the data. Variant exists which is slightly more complex where the projection matrix (eigenvectors/values) are computed on, a with the (lvq) prototypes, augmented version of the data. Implementation wise the difference is: for the simple variant the projection matrix stays constant and can be provided at the start. The second option requires that the projection is calculated at every update step.
Paper describing the idea:
Strickert, M., Hammer, B., Villmann, T., & Biehl, M. (2013). Regularization and improved interpretation of linear data mappings and adaptive distance measures. Proceedings of the 2013 IEEE Symposium on Computational Intelligence and Data Mining, CIDM 2013 - 2013 IEEE Symposium Series on Computational Intelligence, SSCI 2013, 10–17. https://doi.org/10.1109/CIDM.2013.6597211
Is your feature request related to a problem? Please describe. Not really a problem but a nice feature to have.
Describe the solution you'd like Add the null-space correction, i.e., project out the null-space of the covariance matrix of the data. Variant exists which is slightly more complex where the projection matrix (eigenvectors/values) are computed on, a with the (lvq) prototypes, augmented version of the data. Implementation wise the difference is: for the simple variant the projection matrix stays constant and can be provided at the start. The second option requires that the projection is calculated at every update step.
Paper describing the idea:
Describe alternatives you've considered N/a
Additional context N/a