Expression tree API like CVXPY

[carlosgmartin]

Does jaxopt have an expression tree API like CVXPY does?

If not, would it be possible to create one? This would make it easier to build up problems.

[vroulet]

Hello @carlosgmartin, Wouldn't jaxprs be what you are looking for? JAX traces the whole computational graph so that you could directly work on it. I may not have got what you are looking for. Feel free to detail more your objective.

[carlosgmartin]

@vroulet I meant something like the example shown in the second link (more examples here). Here's another example:

def find_nash_equilibrium(u):
    """Find the Nash equilibrium of a two-player zero-sum normal-form game.
    u is the payoff matrix for the row player."""
    x = cp.Variable(u.shape[0])
    v = cp.Variable()
    objective = cp.Maximize(v)
    constraints = [
        v <= x @ u,
        x >= 0,
        x.sum() == 1,
    ]
    problem = cp.Problem(objective, constraints)
    result = problem.solve()
    return {"v": v.value, "x": x.value, "y": constraints[0].dual_value}

That is, letting users create variables and build up expressions from them to create objectives and constraints for a desired problem. This makes it easier for users to write linear/quadratic programs than manually fiddling with the A, b, G, h, Q, c arrays.

[vroulet]

So no, we haven't that now and we were not planning on adding it. Wouldn't a package like https://github.com/cvxgrp/cvxpylayers be a good starting point?