Ensure fitresult.covar is not None before processing

[martinvonk]

I'm running into an issue where the fitresult.covar is None.

Fit succeeded. Could not estimate error-bars.
[[Fit Statistics]]
    # fitting method   = leastsq
    # function evals   = 34
    # data points      = 300
    # variables        = 3
    chi-square         = 22518.4864
    reduced chi-square = 75.8198197
    Akaike info crit   = 1301.49282
    Bayesian info crit = 1312.60417
##  Warning: uncertainties could not be estimated:
    c_0_0:    at initial value
[[Variables]]
    kaq_0_1:  256.584548 (init = 100)
    c_0_0:    1000.00000 (init = 1000)
    Saq_0_1:  7.4078e-05 (init = 0.0001)

[mbakker7]

Thanks for reporting this. What happens right now when the covar cannot be computed? Does it bomb or does it simply not compute stderr.
We are in the process of updating this internally. In fact, the current plan is to not compute stderr at all, as it rarely is a reasonable estimate since the residuals are highly correlated.

[martinvonk]

It now just crashes on np.sqrt(np.diag(self.fitresult.covar)) because self.fitresult.covar is None. So I think it would be nice to at least solve this to prevent crashes. You can decide on showing the stderr or not for cases that don't crash :)

See below:

Cell In[3], [line 53](vscode-notebook-cell:?execution_count=3&line=53)
     52 cal.series(name="obs", x=x, y=y, t=t_obs, h=h_obs, layer=0)
---> [53](vscode-notebook-cell:?execution_count=3&line=53) cal.fit(report=False)

File ~.venv/lib/python3.13/site-packages/timflow/transient/fit.py:391, in Calibrate.fit(self, report, printdot, **kwargs)
    388 def fit(self, report=False, printdot=True, **kwargs):
    389     # current default fitting routine is lmfit
    390     # return self.fit_least_squares(report) # does not support bounds by default
--> [391](.venv/lib/python3.13/site-packages/timflow/transient/fit.py:391)     return self.fit_lmfit(report, printdot, **kwargs)

File ~.venv/lib/python3.13/site-packages/timflow/transient/fit.py:348, in Calibrate.fit_lmfit(self, report, printdot, **kwargs)
    344     self.parameters.loc[name, "optimal"] = self.fitresult.params.valuesdict()[
    345         name
    346     ]
    347 if hasattr(self.fitresult, "covar"):
--> [348](.venv/lib/python3.13/site-packages/timflow/transient/fit.py:348)     self.parameters["std"] = np.sqrt(np.diag(self.fitresult.covar))
    349     self.parameters["perc_std"] = (
    350         100 * self.parameters["std"] / np.abs(self.parameters["optimal"])
    351     )
    352 else:

File ~/.venv/lib/python3.13/site-packages/numpy/lib/_twodim_base_impl.py:328, in diag(v, k)
    326     return diagonal(v, k)
    327 else:
--> [328](~.venv/lib/python3.13/site-packages/numpy/lib/_twodim_base_impl.py:328)     raise ValueError("Input must be 1- or 2-d.")

ValueError: Input must be 1- or 2-d.

[mbakker7]

Well that is no good. So your fix is probably needed until we move forward on the stderr issue.

[dbrakenhoff]

I removed the parameter uncertainty stuff in the unified calibration class. We could consider just removing the computation of that stuff here as well? Otherwise the proposed fix seems fine for now.

[mbakker7]

This PR is replaced by #105.