Predictive Linear Dimension Reduction
PBLDR is a dimensionality reduction method, whose objective is to produce observable linear trends of the input variables in the lower dimensional representation. Moreover, considers the existence of a set of prediction variables, which are also fitted to be linearly trended. The code produces two solutions, one analytical and one numerical. The analytical tends to have higher prediction error than the numerical, due to some instability in the solution.
The repository data from the paper Muñoz, et al. "Instance Spaces for Machine Learning Classification" Accepted in the Machine Learning Journal, which is available at https://www.researchgate.net/publication/315835025_Instance_Spaces_for_Machine_Learning_Classification