Dr. Lary wrote code to reduce the size of the training set as a pre-processing step before Gaussian Process Regression/Classification. Since the exact algorithm is $\mathcal{O}(n^3)$, things get slow very fast. To deal with this, the idea was to first compute 100 quantiles for each feature as well as the target. Finally, we sample the rows of our dataset and return a subset with a more uniform distribution .
Dr. Lary wrote code to reduce the size of the training set as a pre-processing step before Gaussian Process Regression/Classification. Since the exact algorithm is $\mathcal{O}(n^3)$, things get slow very fast. To deal with this, the idea was to first compute 100 quantiles for each feature as well as the target. Finally, we sample the rows of our dataset and return a subset with a more uniform distribution .