PRs #6 and #8 provide improvements to quantile_ensemble_flex performance by speeding up the formation of the constraint matrix fed to the LP solver. The same type of approach might be applicable to the non-ensemble quantile regression code.
Tasks include:
Benchmarking current implementation (see if matrix formation is a significant contributor to runtime)
Preparing characterization tests to catch any inadvertent changes to behavior
Implementing, testing, and benchmarking alternative constraint matrix formation approaches, probably working off of #8's approach (listTsparse-from-parts; the tau group handling does not apply here).
PRs #6 and #8 provide improvements to
quantile_ensemble_flexperformance by speeding up the formation of the constraint matrix fed to the LP solver. The same type of approach might be applicable to the non-ensemble quantile regression code.Tasks include: