Dimension Reduction Benchmark

Dimension reduction benchmark (gKDR)

Plot L_1 model error against reduced dimension, for a one-hundred dimensional problem. The model is a Gaussian process (with maximum-likelihood hyperparameters), with inputs determined by gKDR, varying structural dimension and with a selection of scale parameters. The inputs to the model are 100 points chosen uniformly at random in 100 dimensions, and the observations are a linear mapping of these into one dimension.