formatting

templatefitter/utility.py

@@ -4,8 +4,13 @@
 from numba import jit, vectorize, float64, float32
 
 
-__all__ = ["cov2corr", "corr2cov", "xlogyx", "get_systematic_cov_mat",
-           "array_split_into"]
+__all__ = [
+    "cov2corr",
+    "corr2cov",
+    "xlogyx",
+    "get_systematic_cov_mat",
+    "array_split_into",
+]
 
 
 def cov2corr(cov):
@@ -77,19 +82,23 @@ def id_to_index(names, param_id):
 
 
 # @jit(nopython=True, cache=True)
-@vectorize([float32(float32, float32),
-            float64(float64, float64)])
+@vectorize([float32(float32, float32), float64(float64, float64)])
 def xlogyx(x, y):
     """Compute :math:`x*log(y/x)`to a good precision when :math:`y~x`.
     The xlogyx function is taken from https://github.com/scikit-hep/probfit/blob/master/probfit/_libstat.pyx.
     """
     # result = np.where(x < y, x * np.log1p((y - x) / x), -x * np.log1p((x - y) / y))
-    if x < 1e-100:
-        return 0.
-    elif x<y:
-        return x*np.log1p((y-x)/x)
+
+    if np.isnan(x):
+        return 0.0
+
+    elif x < 1e-100:
+        return 0.0
+
+    elif x < y:
+        return x * np.log1p((y - x) / x)
     else:
-        return -x*np.log1p((x-y)/y)
+        return -x * np.log1p((x - y) / y)
     # result = np.zeros_like(x)
     #
     # for i in range(x.shape[0]):