Merged in bugfix/RAM-1713_pf_small_sets (pull request #351)

RAM-1713 Fix required prominence for small picket fences

Approved-by: Randy Taylor

docs/source/changelog.rst

@@ -36,6 +36,16 @@ CT
   The slice spacing distance will now use the distance between ``ImagePositionPatient`` tags of
   the first two slices to avoid reliance on the ``SpacingBetweenSlices`` tag.
 
+Picket Fence
+^^^^^^^^^^^^
+
+* Picket fences where only a relatively small subset of the leaves were being analyzed (e.g. 10 pairs) were sometimes failing.
+  This would produce a ``ValueError: cannot convert float NaN to integer`` error. This has been fixed.
+  As a workaround, often the fix was to set ``required_prominence`` to a small value or None. This is no longer necessary.
+  ``required_prominence`` now reflects the *normalized* height (0-1.0) the pickets should be above the background. Previously,
+  this value was not normalized, requiring fiddling with the value to get correct and depending on the number
+  of leaf pairs that were being analyzed. The number of leaf pairs should no longer be a factor in the analysis.
+
 Image Metrics
 ^^^^^^^^^^^^^
 

pylinac/picketfence.py

@@ -680,6 +680,7 @@ def analyze(
         else:
             leaf_prof = np.mean(self.image, 1)
         leaf_prof = MultiProfile(leaf_prof)
+        leaf_prof.normalize()  # normalize so required prominence is also normalized.
         peak_idxs, peak_vals = leaf_prof.find_fwxm_peaks(
             min_distance=0.02,
             threshold=height_threshold,

tests_basic/test_picketfence.py

@@ -957,3 +957,13 @@ class CharlestonRA(PFTestMixin, TestCase):
 class CharlestonG0(PFTestMixin, TestCase):
     file_name = ["Charleston", "TB1", "July2016", "G0.dcm"]
     max_error = 0.1
+
+
+class CanberraShortSet(PFTestMixin, TestCase):
+    """This is a small picket set (~10-20 leaves). Should be no problem"""
+
+    file_name = "canberra.dcm"
+    max_error = 0
+    abs_median_error = 0
+    num_pickets = 10
+    mean_picket_spacing = 15