I'm trying to build a model using lmfit (link to docs) and I can't seems to find out why I keep getting a ValueError: The input contains nan values when I try to fit the model.
from lmfit import minimize, Minimizer, Parameters, Parameter, report_fit, Model import numpy as np def cde(t, Qi, at, vw, R, rhob_cb, al, d, r): # t (time), is the independent variable return Qi / (8 * np.pi * ((at * vw)/R) * t * rhob_cb * (np.sqrt(np.pi * ((al * vw)/R * t)))) * \ np.exp(- (R * (d - (t * vw)/ R)**2) / (4 * (al * vw) * t) - (R * r**2)/ (4 * (at * vw) * t)) model_cde = Model(cde) # Allowed to vary model_cde.set_param_hint('vw', value =10**-4, min=0.000001) model_cde.set_param_hint('d', value = -0.038, min = 0.0001) model_cde.set_param_hint('r', value = 5.637e-10) model_cde.set_param_hint('at', value =0.1) model_cde.set_param_hint('al', value =0.15) # Fixed model_cde.set_param_hint('Qi', value = 1000, vary = False) model_cde.set_param_hint('R', value =1.7, vary = False) model_cde.set_param_hint('rhob_cb', value =3000, vary = False) # test data data = [ 1.37, 1.51, 1.65, 1.79, 1.91, 2.02, 2.12, 2.2 , 2.27, 2.32, 2.36, 2.38, 2.4 , 2.41, 2.42, 2.41, 2.4 , 2.39, 2.37, 2.35, 2.33, 2.31, 2.29, 2.26, 2.23, 2.2 , 2.17, 2.14, 2.11, 2.08, 2.06, 2.02, 1.99, 1.97, 1.94, 1.91, 1.88, 1.85, 1.83, 1.8 , 1.78, 1.75, 1.72, 1.7 , 1.68, 1.65, 1.63, 1.61, 1.58] time = list(range(5,250,5)) model_cde.fit(data, t= time) Produces the following error:
--------------------------------------------------------------------------- ValueError Traceback (most recent call last) <ipython-input-16-785fcc6a994b> in <module>() ----> 1 model_cde.fit(data, t= time) /home/bprodz/.virtualenvs/phd_dev/lib/python3.5/site-packages/lmfit/model.py in fit(self, data, params, weights, method, iter_cb, scale_covar, verbose, fit_kws, **kwargs) 539 scale_covar=scale_covar, fcn_kws=kwargs, 540 **fit_kws) --> 541 output.fit(data=data, weights=weights) 542 output.components = self.components 543 return output /home/bprodz/.virtualenvs/phd_dev/lib/python3.5/site-packages/lmfit/model.py in fit(self, data, params, weights, method, **kwargs) 745 self.init_fit = self.model.eval(params=self.params, **self.userkws) 746 --> 747 _ret = self.minimize(method=self.method) 748 749 for attr in dir(_ret): /home/bprodz/.virtualenvs/phd_dev/lib/python3.5/site-packages/lmfit/minimizer.py in minimize(self, method, params, **kws) 1240 val.lower().startswith(user_method)): 1241 kwargs['method'] = val -> 1242 return function(**kwargs) 1243 1244 /home/bprodz/.virtualenvs/phd_dev/lib/python3.5/site-packages/lmfit/minimizer.py in leastsq(self, params, **kws) 1070 np.seterr(all='ignore') 1071 -> 1072 lsout = scipy_leastsq(self.__residual, vars, **lskws) 1073 _best, _cov, infodict, errmsg, ier = lsout 1074 result.aborted = self._abort /home/bprodz/.virtualenvs/phd_dev/lib/python3.5/site-packages/scipy/optimize/minpack.py in leastsq(func, x0, args, Dfun, full_output, col_deriv, ftol, xtol, gtol, maxfev, epsfcn, factor, diag) 385 maxfev = 200*(n + 1) 386 retval = _minpack._lmdif(func, x0, args, full_output, ftol, xtol, --> 387 gtol, maxfev, epsfcn, factor, diag) 388 else: 389 if col_deriv: /home/bprodz/.virtualenvs/phd_dev/lib/python3.5/site-packages/lmfit/minimizer.py in __residual(self, fvars, apply_bounds_transformation) 369 370 out = self.userfcn(params, *self.userargs, **self.userkws) --> 371 out = _nan_policy(out, nan_policy=self.nan_policy) 372 373 if callable(self.iter_cb): /home/bprodz/.virtualenvs/phd_dev/lib/python3.5/site-packages/lmfit/minimizer.py in _nan_policy(a, nan_policy, handle_inf) 1430 1431 if contains_nan: -> 1432 raise ValueError("The input contains nan values") 1433 return a 1434 ValueError: The input contains nan values However the results of the following checks for NaNs confirms that there were no NaN values in my data:
print(np.any(np.isnan(data)), np.any(np.isnan(time))) False False So far I've tried converting 1 and/or both of data and time from lists to numpy ndarrays, removing the 0th time step (in case there was a dividing by 0 error), explicitly specifying the t as being independent and allowing all variables to vary. However these all throw the same error.
Does anyone have ideas what is causing this error to be thrown? Thanks.
61 Answer
I tried to fit my model using scipy.optimize.curve_fit and got the following error:
/home/bprodz/.virtualenvs/phd_dev/lib/python3.4/site-packages/ipykernel/__main__.py:3: RuntimeWarning: invalid value encountered in sqrt app.launch_new_instance() Which suggests the problem is with my model generating some negative numbers for np.sqrt(). The default behaviour for np.sqrt() when given a negative number is to output nan as per this question. NB the np.sqrt can be set to raise an error if given a negative number be setting the following: np.seterr(all='raise') source
TIP I also asked for help in the lmfit google group and received the following helpful advice:
- Consider breaking long formulas into smaller pieces to make troubleshooting easier
- Use
Model.eval()to test what certain parameters will produce when run through your model function np.ndarrayis generally superior to python lists in these (numerical) situations
