Fix ValueError when refitting after load(fitted_as_initial=True)

[gaoflow]

Summary

Loading a previously-fitted model with fitted_as_initial=True and then refitting it on a new spectrum raises a ValueError, as reported in #310:

ValueError: operands could not be broadcast together with shapes (n,) (0,)

Root cause

BaseCircuit.load(..., fitted_as_initial=True) stores the fitted parameters as a numpy array:

self.initial_guess = np.array(json_data['Parameters'])

The subsequent fit() then checks emptiness with if self.initial_guess != []:. For a numpy array this is an element-wise comparison, which NumPy tries to broadcast against the empty list [] (shape (0,)) and raises the broadcasting error. (This matches the workaround noted in the issue thread — manually casting initial_guess back to a list.)

Fix

Test for emptiness with len(self.initial_guess) != 0 instead. This is robust whether initial_guess is a list (the construction-time type) or a numpy array (the post-load type), and preserves the existing behaviour for the empty case.

Reproduction

import numpy as np
from impedance.models.circuits import CustomCircuit

c = CustomCircuit('R0-p(R1,C1)', initial_guess=[90, 180, 2e-5])
c.fit(freqs, Z)
c.save('m.json')

c2 = CustomCircuit()
c2.load('m.json', fitted_as_initial=True)   # initial_guess is now an ndarray
c2.fit(freqs, Z)                            # -> ValueError before this fix

Verification

Added a regression test test_refit_after_load_fitted_as_initial in impedance/tests/test_model_io.py that fits a model, saves it, reloads it with fitted_as_initial=True, and refits. It fails on the current code with the reported (n,) (0,) broadcasting error and passes with this fix. The existing test_model_io suite continues to pass.

Fixes #310.