Although the code runs properly, a warning is raised where we assign complex values to real variables.
These complex values are nothing less than the partial derivatives of the cost function with respect to the angles in the circuit. It's not difficult to see that there must be something wrong behind. Probably the expressions we have to analytically obtain these derivatives are wrong.

The expressions we are using are:

Where both and are complex.

Since everything comes from the chain rule, the imaginary parts of both should cancel, and in the end render purely real numbers, which would compose the gradient.

I repeat, that does not happen.

It would be great to add some aid to outcomers.
Either an example notebook or a test script where one sees how the different parts of the code are used would ease things out a lot when asking for help or when presenting our work.

cya!