Merge ad7124e into f0958fe

ccdproc/ccdproc.py

@@ -178,17 +178,23 @@ def subtract_overscan(ccd, overscan=None, fits_section=None,
     if fits_section is not None:
         overscan = ccd[slice_from_string(fits_section, fits_convention=True)]
 
+    # Assume axis with the smaller dimension is the one to aggregate over
+    oscan_axis = 1 if overscan.shape[0] > overscan.shape[1] else 0
+
     if median:
-        oscan = np.median(overscan.data, axis=1)
+        oscan = np.median(overscan.data, axis=oscan_axis)
     else:
-        oscan = np.mean(overscan.data, axis=1)
+        oscan = np.mean(overscan.data, axis=oscan_axis)
 
     if model is not None:
         of = fitting.LinearLSQFitter()
         yarr = np.arange(len(oscan))
         oscan = of(model, yarr, oscan)
         oscan = oscan(yarr)
-        oscan = np.reshape(oscan, (oscan.size, 1))
+        if oscan_axis == 1:
+            oscan = np.reshape(oscan, (oscan.size, 1))
+        else:
+            oscan = np.reshape(oscan, (1, oscan.size))
     else:
         oscan = np.reshape(oscan, oscan.shape + (1,))
 

ccdproc/tests/test_ccdproc.py

@@ -69,15 +69,21 @@ def test_create_variance_keywords_must_have_unit(ccd_data):
 
 
 # tests for overscan
-@pytest.mark.parametrize('median,model', [
-                         (False, None),
-                         (True, None), ])
-def test_subtract_overscan_mean(ccd_data, median, model):
+@pytest.mark.parametrize('median,transpose', [
+                         (False, False),
+                         (False, True),
+                         (True, False), ])
+def test_subtract_overscan(ccd_data, median, transpose):
     # create the overscan region
     oscan = 300.
     oscan_region = (slice(None), slice(0, 10))  # indices 0 through 9
     fits_section = '[1:10, :]'
     science_region = (slice(None), slice(10, None))
+    if transpose:
+        # Put overscan in first axis, not second, a test for #70
+        oscan_region = oscan_region[::-1]
+        fits_section = '[:, 1:10]'
+        science_region = science_region[::-1]
     ccd_data.data[oscan_region] = oscan
     # Add a fake sky background so the "science" part of the image has a
     # different average than the "overscan" part.
@@ -86,8 +92,8 @@ def test_subtract_overscan_mean(ccd_data, median, model):
     ccd_data.data[science_region] += oscan + sky
     # Test once using the overscan argument to specify the overscan region
     ccd_data_overscan = subtract_overscan(ccd_data,
-                                          overscan=ccd_data[:, 0:10],
-                                          median=median, model=model)
+                                          overscan=ccd_data[oscan_region],
+                                          median=median, model=None)
     # Is the mean of the "science" region the sum of sky and the mean the
     # "science" section had before backgrounds were added?
     np.testing.assert_almost_equal(
@@ -100,7 +106,7 @@ def test_subtract_overscan_mean(ccd_data, median, model):
     # with the fits_section
     ccd_data_fits_section = subtract_overscan(ccd_data,
                                               fits_section=fits_section,
-                                              median=median, model=model)
+                                              median=median, model=None)
     # Is the mean of the "science" region the sum of sky and the mean the
     # "science" section had before backgrounds were added?
     np.testing.assert_almost_equal(
@@ -115,14 +121,34 @@ def test_subtract_overscan_mean(ccd_data, median, model):
 
 
 # A more substantial test of overscan modeling
-def test_subtract_overscan_model(ccd_data):
+@pytest.mark.parametrize('transpose', [
+                         True,
+                         False])
+def test_subtract_overscan_model(ccd_data, transpose):
     # create the overscan region
     size = ccd_data.shape[0]
+
+    oscan_region = (slice(None), slice(0, 10))
+    science_region = (slice(None), slice(10, None))
+
     yscan, xscan = np.mgrid[0:size, 0:size] / 10.0 + 300.0
-    ccd_data.data = ccd_data.data + yscan
-    ccd_data = subtract_overscan(ccd_data, overscan=ccd_data[:, 0:10],
+
+    if transpose:
+        oscan_region = oscan_region[::-1]
+        science_region = science_region[::-1]
+        scan = xscan
+    else:
+        scan = yscan
+
+    original_mean = ccd_data.data[science_region].mean()
+
+    ccd_data.data[oscan_region] = 0.  # only want overscan in that region
+    ccd_data.data = ccd_data.data + scan
+
+    ccd_data = subtract_overscan(ccd_data, overscan=ccd_data[oscan_region],
                                  median=False, model=models.Polynomial1D(2))
-    assert abs(ccd_data.data.mean()) < 0.1
+    np.testing.assert_almost_equal(ccd_data.data[science_region].mean(),
+                                   original_mean)
 
 
 def test_subtract_overscan_fails(ccd_data):