Numpy-dev failure for modelling #7678

�[1m�[31m/home/travis/miniconda/envs/test/lib/python3.6/site-packages/scipy/optimize/minpack.py�[0m:394: TypeError
_____________ ERROR at setup of TestJointFitter.test_joint_fitter ______________

self = <class 'astropy.modeling.tests.test_fitters.TestJointFitter'>

�[1m    def setup_class(self):�[0m
�[1m        """�[0m
�[1m            Create 2 gaussian models and some data with noise.�[0m
�[1m            Create a fitter for the two models keeping the amplitude parameter�[0m
�[1m            common for the two models.�[0m
�[1m            """�[0m
�[1m        self.g1 = models.Gaussian1D(10, mean=14.9, stddev=.3)�[0m
�[1m        self.g2 = models.Gaussian1D(10, mean=13, stddev=.4)�[0m
�[1m        self.jf = JointFitter([self.g1, self.g2],�[0m
�[1m                                      {self.g1: ['amplitude'],�[0m
�[1m                                       self.g2: ['amplitude']}, [9.8])�[0m
�[1m        self.x = np.arange(10, 20, .1)�[0m
�[1m        y1 = self.g1(self.x)�[0m
�[1m        y2 = self.g2(self.x)�[0m
�[1m    �[0m
�[1m        with NumpyRNGContext(_RANDOM_SEED):�[0m
�[1m            n = np.random.randn(100)�[0m
�[1m    �[0m
�[1m        self.ny1 = y1 + 2 * n�[0m
�[1m        self.ny2 = y2 + 2 * n�[0m
�[1m>       self.jf(self.x, self.ny1, self.x, self.ny2)�[0m

�[1m�[31mastropy/modeling/tests/test_fitters.py�[0m:155: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
�[1m�[31mastropy/modeling/fitting.py�[0m:1219: in __call__
�[1m    self.fitparams, args=args)�[0m
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

func = <bound method JointFitter.objective_function of <astropy.modeling.fitting.JointFitter object at 0x7fe06e88cdd8>>
x0 = array([9.8, array([14.9]), array([0.3]), array([13.]), array([0.4])],
      dtype=object)
args = (array([10. , 10.1, 10.2, 10.3, 10.4, 10.5, 10.6, 10.7, 10.8, 10.9, 11. ,
       11.1, 11.2, 11.3, 11.4, 11.5, 11.6, 1...595e+00, -1.30030932e+00, -2.36494839e+00,
       -2.25212374e+00,  3.26000074e+00,  1.37397670e+00,  4.73034211e-02]))
Dfun = None, full_output = 0, col_deriv = 0, ftol = 1.49012e-08
xtol = 1.49012e-08, gtol = 0.0, maxfev = 1200, epsfcn = 2.220446049250313e-16
factor = 100, diag = None

�[1m    def leastsq(func, x0, args=(), Dfun=None, full_output=0,�[0m
�[1m                col_deriv=0, ftol=1.49012e-8, xtol=1.49012e-8,�[0m
�[1m                gtol=0.0, maxfev=0, epsfcn=None, factor=100, diag=None):�[0m
�[1m        """�[0m
�[1m        Minimize the sum of squares of a set of equations.�[0m
�[1m    �[0m
�[1m        ::�[0m
�[1m    �[0m
�[1m            x = arg min(sum(func(y)**2,axis=0))�[0m
�[1m                     y�[0m
�[1m    �[0m
�[1m        Parameters�[0m
�[1m        ----------�[0m
�[1m        func : callable�[0m
�[1m            should take at least one (possibly length N vector) argument and�[0m
�[1m            returns M floating point numbers. It must not return NaNs or�[0m
�[1m            fitting might fail.�[0m
�[1m        x0 : ndarray�[0m
�[1m            The starting estimate for the minimization.�[0m
�[1m        args : tuple, optional�[0m
�[1m            Any extra arguments to func are placed in this tuple.�[0m
�[1m        Dfun : callable, optional�[0m
�[1m            A function or method to compute the Jacobian of func with derivatives�[0m
�[1m            across the rows. If this is None, the Jacobian will be estimated.�[0m
�[1m        full_output : bool, optional�[0m
�[1m            non-zero to return all optional outputs.�[0m
�[1m        col_deriv : bool, optional�[0m
�[1m            non-zero to specify that the Jacobian function computes derivatives�[0m
�[1m            down the columns (faster, because there is no transpose operation).�[0m
�[1m        ftol : float, optional�[0m
�[1m            Relative error desired in the sum of squares.�[0m
�[1m        xtol : float, optional�[0m
�[1m            Relative error desired in the approximate solution.�[0m
�[1m        gtol : float, optional�[0m
�[1m            Orthogonality desired between the function vector and the columns of�[0m
�[1m            the Jacobian.�[0m
�[1m        maxfev : int, optional�[0m
�[1m            The maximum number of calls to the function. If `Dfun` is provided�[0m
�[1m            then the default `maxfev` is 100*(N+1) where N is the number of elements�[0m
�[1m            in x0, otherwise the default `maxfev` is 200*(N+1).�[0m
�[1m        epsfcn : float, optional�[0m
�[1m            A variable used in determining a suitable step length for the forward-�[0m
�[1m            difference approximation of the Jacobian (for Dfun=None).�[0m
�[1m            Normally the actual step length will be sqrt(epsfcn)*x�[0m
�[1m            If epsfcn is less than the machine precision, it is assumed that the�[0m
�[1m            relative errors are of the order of the machine precision.�[0m
�[1m        factor : float, optional�[0m
�[1m            A parameter determining the initial step bound�[0m
�[1m            (``factor * || diag * x||``). Should be in interval ``(0.1, 100)``.�[0m
�[1m        diag : sequence, optional�[0m
�[1m            N positive entries that serve as a scale factors for the variables.�[0m
�[1m    �[0m
�[1m        Returns�[0m
�[1m        -------�[0m
�[1m        x : ndarray�[0m
�[1m            The solution (or the result of the last iteration for an unsuccessful�[0m
�[1m            call).�[0m
�[1m        cov_x : ndarray�[0m
�[1m            Uses the fjac and ipvt optional outputs to construct an�[0m
�[1m            estimate of the jacobian around the solution. None if a�[0m
�[1m            singular matrix encountered (indicates very flat curvature in�[0m
�[1m            some direction).  This matrix must be multiplied by the�[0m
�[1m            residual variance to get the covariance of the�[0m
�[1m            parameter estimates -- see curve_fit.�[0m
�[1m        infodict : dict�[0m
�[1m            a dictionary of optional outputs with the key s:�[0m
�[1m    �[0m
�[1m            ``nfev``�[0m
�[1m                The number of function calls�[0m
�[1m            ``fvec``�[0m
�[1m                The function evaluated at the output�[0m
�[1m            ``fjac``�[0m
�[1m                A permutation of the R matrix of a QR�[0m
�[1m                factorization of the final approximate�[0m
�[1m                Jacobian matrix, stored column wise.�[0m
�[1m                Together with ipvt, the covariance of the�[0m
�[1m                estimate can be approximated.�[0m
�[1m            ``ipvt``�[0m
�[1m                An integer array of length N which defines�[0m
�[1m                a permutation matrix, p, such that�[0m
�[1m                fjac*p = q*r, where r is upper triangular�[0m
�[1m                with diagonal elements of nonincreasing�[0m
�[1m                magnitude. Column j of p is column ipvt(j)�[0m
�[1m                of the identity matrix.�[0m
�[1m            ``qtf``�[0m
�[1m                The vector (transpose(q) * fvec).�[0m
�[1m    �[0m
�[1m        mesg : str�[0m
�[1m            A string message giving information about the cause of failure.�[0m
�[1m        ier : int�[0m
�[1m            An integer flag.  If it is equal to 1, 2, 3 or 4, the solution was�[0m
�[1m            found.  Otherwise, the solution was not found. In either case, the�[0m
�[1m            optional output variable 'mesg' gives more information.�[0m
�[1m    �[0m
�[1m        Notes�[0m
�[1m        -----�[0m
�[1m        "leastsq" is a wrapper around MINPACK's lmdif and lmder algorithms.�[0m
�[1m    �[0m
�[1m        cov_x is a Jacobian approximation to the Hessian of the least squares�[0m
�[1m        objective function.�[0m
�[1m        This approximation assumes that the objective function is based on the�[0m
�[1m        difference between some observed target data (ydata) and a (non-linear)�[0m
�[1m        function of the parameters `f(xdata, params)` ::�[0m
�[1m    �[0m
�[1m               func(params) = ydata - f(xdata, params)�[0m
�[1m    �[0m
�[1m        so that the objective function is ::�[0m
�[1m    �[0m
�[1m               min   sum((ydata - f(xdata, params))**2, axis=0)�[0m
�[1m             params�[0m
�[1m    �[0m
�[1m        The solution, `x`, is always a 1D array, regardless of the shape of `x0`,�[0m
�[1m        or whether `x0` is a scalar.�[0m
�[1m        """�[0m
�[1m        x0 = asarray(x0).flatten()�[0m
�[1m        n = len(x0)�[0m
�[1m        if not isinstance(args, tuple):�[0m
�[1m            args = (args,)�[0m
�[1m        shape, dtype = _check_func('leastsq', 'func', func, x0, args, n)�[0m
�[1m        m = shape[0]�[0m
�[1m        if n > m:�[0m
�[1m            raise TypeError('Improper input: N=%s must not exceed M=%s' % (n, m))�[0m
�[1m        if epsfcn is None:�[0m
�[1m            epsfcn = finfo(dtype).eps�[0m
�[1m        if Dfun is None:�[0m
�[1m            if maxfev == 0:�[0m
�[1m                maxfev = 200*(n + 1)�[0m
�[1m            with _MINPACK_LOCK:�[0m
�[1m                retval = _minpack._lmdif(func, x0, args, full_output, ftol, xtol,�[0m
�[1m>                                        gtol, maxfev, epsfcn, factor, diag)�[0m
�[1m�[31mE               TypeError: Cannot cast array data from dtype('O') to dtype('float64') according to the rule 'safe'�[0m