transform 0-d arrays into 1-d to check their length (#952)

* transform 0-d arrays into 1-d to check their length

* use recommended solution

* fix tests

lmfit/model.py

@@ -1689,7 +1689,7 @@ def eval_uncertainty(self, params=None, sigma=1, dscale=0.01, **kwargs):
         nvarys = self.nvarys
         # ensure fjac and df2 are correct size if independent var updated by kwargs
         feval = self.model.eval(params, **userkws)
-        ndata = len(feval.view('float64'))        # allows feval to be complex
+        ndata = np.atleast_1d(feval).view('float64').ravel().size  # allows feval to be complex
         covar = self.covar
         if any(p.stderr is None for p in params.values()):
             return np.zeros(ndata)
@@ -1720,8 +1720,8 @@ def eval_uncertainty(self, params=None, sigma=1, dscale=0.01, **kwargs):
 
             pars[pname].value = val0
             for key in fjac:
-                fjac[key][i] = (res1[key].view('float64')
-                                - res2[key].view('float64')) / (2*dval)
+                fjac[key][i] = (np.atleast_1d(res1[key]).view('float64').ravel()
+                                - np.atleast_1d(res2[key]).view('float64').ravel()) / (2*dval)
 
         for i in range(nvarys):
             for j in range(nvarys):
@@ -1738,7 +1738,10 @@ def eval_uncertainty(self, params=None, sigma=1, dscale=0.01, **kwargs):
         # for complex data, convert back to real/imag pairs
         if feval.dtype in ('complex64', 'complex128'):
             for key in fjac:
-                df2[key] = df2[key][0::2] + 1j * df2[key][1::2]
+                df2[key] = df2[key].view(feval.dtype)
+
+        for key in fjac:
+            df2[key] = df2[key].reshape(feval.shape)
 
         df2_total = df2.pop('0')
         self.dely = scale * np.sqrt(df2_total)

tests/test_model_uncertainties.py

@@ -5,6 +5,7 @@
 from numpy.testing import assert_allclose
 
 from lmfit.lineshapes import gaussian
+from lmfit.model import Model
 from lmfit.models import ExponentialModel, GaussianModel, LinearModel
 
 
@@ -127,3 +128,23 @@ def test_component_uncertainties():
     assert result.dely_comps['g2_'].mean() < 1.5
     assert result.dely_comps['bkg_'].mean() > 0.5
     assert result.dely_comps['bkg_'].mean() < 1.5
+
+
+def test_scalar_independent_vars():
+    """Github #951"""
+    mr = LinearModel().fit(data=[1, 2, 4], x=[1, 2, 3])
+
+    assert_allclose(mr.eval_uncertainty(x=1.2), np.array([0.60629034]))
+    assert_allclose(mr.eval_uncertainty(x=np.array(1.2j)), np.array([1.15903153+0.17475665j]))
+
+
+def test_multidim_model():
+    """test models that return a multi-dimension output"""
+    def fit(x, p=2):
+        return np.array([x, p*x])
+    model = Model(fit)
+
+    mr = model.fit(data=np.array([[1, 2, 3], [2, 3, 5]]), x=np.array([1, 2, 3]))
+
+    assert_allclose(mr.eval_uncertainty(x=np.array([3, 4])), np.array([[0, 0], [0.18432744, 0.24576991]]))
+    assert_allclose(mr.eval_uncertainty(x=np.array([3j, 4j])), np.array([[0, 0], [0.13033918+0.13033918j, 0.17378557+0.17378557j]]))