William Zijie Zhang
William Zijie Zhang
It seems like the existing canonicalization file for ``quad_over_lin`` already does something similar? ```py def quad_over_lin_canon(expr, args): # quad_over_lin := sum_{ij} X^2_{ij} / y x = args[0] y = args[1].flatten(order='F')...
> This code is not "vectorized" because t is a scalar (and y as well). Vectorizing quad-over-lin is about allowing y to be a vector (or matrix?). ah right, I...
closing this as it is not really necessary anymore (maybe it will be refactored later on with the backend changes).
@Mettush I am currently working on an extension of CVXPY for the IPOPT solver (which also requires gradient oracles and an initial point). You can find some of the progress...
I made the mistake of branching from the docs prep 1.7.0 PR, so we can disregard the docs changes.
closing this as it is out of date and doesn't really work.
@carlosgmartin it seems like the issue comes from the solvers being incompatible with NumPy 2.0? Can you try uninstalling them (ECOS for example) to see if this fixes the issue?
@carlosgmartin I also recently had some issues with NumPy 2.0 compatibilities but noticed that it was because I inadvertly messed up my conda environment. I would suggest you to start...
hmm not sure, maybe @mlubin would know?
[gpkit](https://gpkit.readthedocs.io/en/latest/debugging.html) also has great tools for debugging infeasibility. @langestefan I would be happy to work with you on ways to go about this :). I think taking ideas from JuMP...