Added warnings for normalization and fixed tests

astropy/nddata/convolution/core.py

@@ -14,11 +14,13 @@
 
 Currently only symmetric 2D kernels are supported.
 """
+import copy
+import warnings
 
 import numpy as np
-import copy
+
 from .utils import (discretize_model, add_kernel_arrays_1D,
-                    add_kernel_arrays_2D, NormalizationWarning)
+                    add_kernel_arrays_2D)
 
 
 class Kernel(object):
@@ -33,8 +35,11 @@ class Kernel(object):
     def __init__(self, array):
         self._array = array
         self._normalization = 1. / self._array.sum()
-        if np.abs(self._normalization) > self._array.size * self._normalization:
-            raise NormalizationWarning("Normalization factor of kernel is" +
+        # The value of 100 is kind of arbitrary
+        # there are kernel sum to zero and the user should
+        # be warned in this case
+        if np.abs(self._normalization) > 100:
+            warnings.warn("Normalization factor of kernel is" +
                                         "exceptionally large.")
 
     @property

astropy/nddata/convolution/kernels.py

@@ -1,18 +1,19 @@
 from __future__ import division
+import warnings
+
 import numpy as np
 
 from .core import Kernel1D, Kernel2D, Kernel
 from .utils import KernelSizeError
-
 from ...modeling.functional_models import *
 from ...modeling.core import Parametric1DModel, Parametric2DModel
 
-
 __all__ = sorted(['Gaussian1DKernel', 'Gaussian2DKernel', 'CustomKernel',
                   'Box1DKernel', 'Box2DKernel', 'Tophat2DKernel',
                   'Trapezoid1DKernel', 'MexicanHat1DKernel',
                   'MexicanHat2DKernel', 'AiryDisk2DKernel',
-                  'Model1DKernel', 'Model2DKernel'])
+                  'Model1DKernel', 'Model2DKernel',
+                  'TrapezoidDisk2DKernel'])
 
 
 class Gaussian1DKernel(Kernel1D):
@@ -212,7 +213,7 @@ class TrapezoidDisk2DKernel(Kernel2D):
     _weighted = True
 
     def __init__(self, width, slope=1., **kwargs):
-        self._model = TrapezoidDisk2DModel(1, 0, width, slope)
+        self._model = TrapezoidDisk2DModel(1, 0, 0, width, slope)
         self._default_size = width + 2. / slope
         super(TrapezoidDisk2DKernel, self).__init__(**kwargs)
         self._truncation = 0
@@ -242,7 +243,9 @@ class MexicanHat1DKernel(Kernel1D):
     def __init__(self, width, **kwargs):
         self._default_size = 8 * width
         self._model = MexicanHat1DModel(1, 0, width)
-        super(MexicanHat1DKernel, self).__init__(**kwargs)
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore")
+            super(MexicanHat1DKernel, self).__init__(**kwargs)
         self._truncation = np.abs(self._array.sum() / self._array.size)
         self._normalization = 0
 
@@ -274,7 +277,9 @@ class MexicanHat2DKernel(Kernel2D):
     def __init__(self, width, **kwargs):
         self._default_size = 8 * width
         self._model = MexicanHat2DModel(1, 0, 0, width)
-        super(MexicanHat2DKernel, self).__init__(**kwargs)
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore")
+            super(MexicanHat2DKernel, self).__init__(**kwargs)
         self._truncation = np.abs(self._array.sum() / self._array.size)
         self._normalization = 0
 

astropy/nddata/convolution/tests/test_convolve_kernels.py

@@ -11,7 +11,10 @@
 
 
 widths = [3, 5, 7, 9]
-kernel_types = [GaussianKernel, BoxKernel, Tophat2DKernel, MexicanHat1DKernel]
+kernel_types = [Gaussian1DKernel, Gaussian2DKernel,
+                Box1DKernel, Box2DKernel,
+                Trapezoid1DKernel, TrapezoidDisk2DKernel,
+                MexicanHat1DKernel, Tophat2DKernel]
 
 
 class Test2DConvolutions(object):
@@ -80,15 +83,13 @@ def test_smallkernel_vs_makekernel(self, width):
 
         Compares a small kernel to something produced by makekernel
         """
-        box = BoxKernel(width)
-        kernel1 = np.ones([width, width])
-        kernel2 = np.outer(box.array, box.array)
+        kernel1 = np.ones([width, width]) / width ** 2
+        kernel2 = Box2DKernel(width)
 
         x = np.zeros(kernel2.shape)
-        xslice = [slice(sh // 2, sh // 2) for sh in kernel2.shape]
-        x[xslice] = 1.0
+        x[kernel2.center] = 1.0
 
-        c2 = convolve_fft(x, kernel2, boundary='fill')
+        c2 = convolve_fft(x, kernel2.array, boundary='fill')
         c1 = convolve_fft(x, kernel1, boundary='fill')
 
         assert_almost_equal(c1, c2, decimal=12)

astropy/nddata/convolution/utils.py

@@ -1,20 +1,10 @@
 from __future__ import division
+
 import numpy as np
 
 from ...modeling.core import Parametric1DModel, Parametric2DModel
 
 
-class NormalizationWarning(Warning):
-    """
-    Called when normalization of kernels is strange.
-    """
-    def __init__(self, message):
-        self._message = message
-
-    def __str__(self):
-        return self._message
-
-
 class DiscretizationError(Exception):
     """
     Called when discretization of models goes wrong.
@@ -235,4 +225,4 @@ def discretize_integrate_2D(model, range_, mode='analytical'):
     if getattr(model, "integral", False):
         raise NotImplementedError("Currently not supported.")
     else:
-        raise NotImplementedError("Currently not supported.")
+        raise NotImplementedError("Currently not supported.")