Jeremiah Willcock

I tried `float32` and `float64` there (using the `jnp.float32(1)` syntax): `float32`: `TypeError: true_fun and false_fun output must have identical types, got ('DIFFERENT ShapedArray(float32[]) vs. ShapedArray(float64[], weak_type=True)', ('ShapedArray(float32[30])', ('ShapedArray(float32[30,30])', 'ShapedArray(float32[1,30])', 'ShapedArray(float64[],...

``` optimizer = jaxopt.BoxOSQP() optimizer.run( params_obj=( jnp.eye(30, dtype=jnp.float32), jnp.ones((30,), dtype=jnp.float32), ), params_eq=jnp.ones((1, 30), dtype=jnp.float32), params_ineq=(jnp.float32(-1), jnp.float32(1)), ) ```

I am using a TPU and my Colab has a large number of other things in it so I can't share it. Did you turn on `float64` in JAX? That...

I want #1 and #2, at least with the option to run other parts of my code in `float64` on the TPU (which is semi-supported). I would like `BoxOSQP` to...

I did get confused on the API, but I would not have expected a crash like that. Would a better error message be possible? Also, does the problem formulation with...

What I was trying to get at is to determine whether that means that arbitrary linear constraints are supported, not just box constraints. Does that mean this class is a...

It would also be nice to be able to provide only the initial solution point but not the dual information. Also, which namespace the `KKTSolution` class is in does not...

It also looks like you can't just fill in the initial point in `KKTSolution`. The `primal` field needs to have two entries, and it looks like the duals are required...

Oh -- there's a method for filling in the parameters and it isn't just internal? Is that documented somewhere?

I assumed that was an alternative to passing the parameters to `run()`, not a way to create them. Maybe there should be an example using `run()` with an `init_params()` call...