deas-mhumhna

Very cool. I will take a look. Parametric simplex is actually a slightly different algorithm that doesn't require a Phase I and is particularly useful in sparse selection. But I...

You mean a sparse matrix package? I'd be interested in helping. Though I was actually referring to LP having its own API, similar to non-constrained optimization. My parametric simplex implementation...

I'm also curious about another thing: in a SQP application, given initial feasibility (another problem altogether) and a decent line search, each sub QP should have 0 as a feasible...

I was just asking why the current initializer can't be used for QP; why does it require refactoring if the only goal is finding a single feasible point? I've been...

I meant actually as asking rather you image `lp` as having its own Method interface similar to those used in optimize and what I see in `exp/tree/master/linsolve`, not necessarily the...

Ok. I've structured my parametric simplex similar to the current simplex algorithm, but with two exported functions: one that has the same form as `Simplex`, and one that can accept...

In the process of writing `parametric`, I also implemented a few scaling algorithms. Scaling can sometimes increase numerical stability. Any interest in adding these kinds of methods? Also, does your...

This could also be implemented more generally by as gradient/update projection. Either as a function that modifies the search direction and limits step size or modifies the final updated location.

True. I imagine this would work best as a function called during backpropagation in order for the unmodified gradients to still be accessible during each passing through the graph. A...

This may be dead but through a minor object-oriented modification of OSQP you can support arbitrary convex spaces for which the support function as well as (projection onto) the recession...