ensure that correct initial params are used when re-fitting a ModelResult (#961)

ensure that correct initial params are used when re-fitting a ModeRresult

Author: newville

lmfit/model.py

@@ -1255,10 +1255,14 @@ def __init__(self, left, right, op, **kws):
         if 'nan_policy' not in kws:
             kws['nan_policy'] = self.left.nan_policy
 
+        # CompositeModel cannot have a prefix.
+        if 'prefix' in kws:
+            warnings.warn("CompositeModel ignores `prefix` argument")
+            kws['prefix'] = ''
+
         def _tmp(self, *args, **kws):
             pass
         Model.__init__(self, _tmp, **kws)
-
         for side in (left, right):
             prefix = side.prefix
             for basename, hint in side.param_hints.items():
@@ -1548,7 +1552,10 @@ def fit(self, data=None, params=None, weights=None, method=None,
         if data is not None:
             self.data = data
         if params is not None:
-            self.init_params = params
+            self.init_params = deepcopy(params)
+        else:
+            self.init_params = deepcopy(self.params)
+
         if weights is not None:
             self.weights = weights
         if method is not None:
@@ -1559,8 +1566,8 @@ def fit(self, data=None, params=None, weights=None, method=None,
         self.ci_out = None
         self.userargs = (self.data, self.weights)
         self.userkws.update(kwargs)
-        self.init_fit = self.model.eval(params=self.params, **self.userkws)
-        _ret = self.minimize(method=self.method)
+        self.init_fit = self.model.eval(params=self.init_params, **self.userkws)
+        _ret = self.minimize(method=self.method, params=self.init_params)
         self.model.post_fit(_ret)
         _ret.params.create_uvars(covar=_ret.covar)
 

tests/test_model.py

@@ -13,7 +13,7 @@
 from lmfit import Model, Parameters, models
 from lmfit.lineshapes import gaussian, lorentzian, step, voigt
 from lmfit.model import get_reducer, propagate_err
-from lmfit.models import GaussianModel, PseudoVoigtModel
+from lmfit.models import GaussianModel, PseudoVoigtModel, QuadraticModel
 
 
 @pytest.fixture()
@@ -1648,3 +1648,52 @@ def test_custom_variadic_model():
     assert result.nfev > 7
     assert_allclose(result.values['c0'], 5.0, 0.02, 0.02, '', True)
     assert_allclose(result.values['c1'], 3.3, 0.02, 0.02, '', True)
+
+
+def test_model_refitting():
+    """Github #960"""
+    np.random.seed(0)
+    x = np.linspace(0, 100, 5001)
+    y = gaussian(x, amplitude=90, center=60, sigma=4) + 30 + 0.3*x - 0.0030*x*x
+    y += np.random.normal(size=5001, scale=0.5)
+
+    model = GaussianModel(prefix='peak_') + QuadraticModel(prefix='bkg_')
+
+    params = model.make_params(bkg_a=0, bkg_b=0, bkg_c=20, peak_amplitude=200,
+                               peak_center=55, peak_sigma=10)
+
+    result = model.fit(y, params, x=x, method='powell')
+    assert result.chisqr > 12000.0
+    assert result.nfev > 500
+    assert result.params['peak_amplitude'].value > 500
+    assert result.params['peak_amplitude'].value < 5000
+    assert result.params['peak_sigma'].value > 10
+    assert result.params['peak_sigma'].value < 100
+
+    # now re-fit with LM
+    result.fit(y, x=x, method='leastsq')
+
+    assert result.nfev > 25
+    assert result.nfev < 200
+    assert result.chisqr < 2000.0
+
+    assert result.params['peak_amplitude'].value > 85
+    assert result.params['peak_amplitude'].value < 95
+    assert result.params['peak_sigma'].value > 3
+    assert result.params['peak_sigma'].value < 5
+
+    # and assert that the initial value are from the Powell result
+    assert result.init_values['peak_amplitude'] > 1500
+    assert result.init_values['peak_sigma'] > 25
+
+    params = model.make_params(bkg_a=0, bkg_b=-.02, bkg_c=26, peak_amplitude=20,
+                               peak_center=62, peak_sigma=3)
+
+    # now re-fit with LM and these new params
+    result.fit(y, params, x=x, method='leastsq')
+
+    # and assert that the initial value are from the Powell result
+    assert result.init_values['peak_amplitude'] > 19
+    assert result.init_values['peak_amplitude'] < 21
+    assert result.init_values['peak_sigma'] > 2
+    assert result.init_values['peak_sigma'] < 4