Follow-up of QuantileRegressor implementation

QuantileRegressor has been implemented in https://github.com/scikit-learn/scikit-learn/pull/9978 However, it was reported to have an issue with large array of data where linear programming solvers require to form huge matrices for the optimization problem.

A potential solution is to have an online solver using partial_fit. However, it is not clear which solvers would be best to solve this issue.

Has proposed by @agramfort https://github.com/scikit-learn/scikit-learn/pull/9978#issuecomment-847159716, we should benchmark the different available solvers before making an implementation choice.

Pinging @RPyElec @avidale @agramfort @lorentzenchr @GaelVaroquaux that were involved in the discussion in the former pull-request.

Tasks:

• Add support for sparse X (relatively easy). #21086
• Propose an algorithm for a partial_fit method, test and benchmark it (difficult task).

[venkyyuvy]

@RPyElec @avidale thinking about it I think it makes more sense to add your stochastic solver to SGGRegressor as it also has the right API for stochastic solvers and it has all the machinery. It’s “just” a matter of adding one more loss in the gradient computations.

Is there any consensus for this? If so I would be interested in working on it.

[BatMrE]

@BatMrE can I take this up ?

Hi @venkyyuvy Yes you can, I am bit packed