SQUASH ME Make lines shorter

scipy/special/_logsumexp.py

@@ -284,8 +284,8 @@ def log_softmax(x, axis=None):
     >>> y
     array([  0., -inf])
 
-    >>> x = np.array([0, np.log(np.finfo(np.float32).smallest_subnormal)], dtype=np.float32)
-
+    >>> subnormal32 = np.finfo(np.float32).smallest_subnormal
+    >>> x = np.array([0, np.log(subnormal32)], dtype=np.float32)
     >>> y = log_softmax(x)
     >>> y
     array([-1.40130e-45, -1.03279e+02], dtype=float32)
@@ -296,8 +296,8 @@ def log_softmax(x, axis=None):
     >>> y
     array([   0.     , -103.27893], dtype=float32)
 
-    >>> x = np.array([0, np.log(np.finfo(np.float64).smallest_subnormal)], dtype=np.float64)
-
+    >>> subnormal64 = np.finfo(np.float64).smallest_subnormal
+    >>> x = np.array([0, np.log(subnormal64)], dtype=np.float64)
     >>> y = log_softmax(x)
     >>> y
     array([-4.9407e-324, -7.4444e+002])

scipy/special/tests/test_log_softmax.py

@@ -10,9 +10,11 @@
     (np.array([1000, 1]), np.array([0, -999])),
     # we shouldn't return zero on the smallest subnormal input
     (np.array([-np.log(np.finfo(np.float32).smallest_subnormal), 0], dtype=np.float32),
-     np.array([float.fromhex('-0x1.00000p-149'), float.fromhex('-0x1.9d1dap+6')], dtype=np.float32)),
+     np.array([float.fromhex('-0x1.00000p-149'), float.fromhex('-0x1.9d1dap+6')],
+              dtype=np.float32)),
     (np.array([-np.log(np.finfo(np.float64).smallest_subnormal), 0], dtype=np.float64),
-     np.array([float.fromhex('-0x0.0000000000001p-1022'), float.fromhex('-0x1.74385446d71c3p+9')], dtype=np.float64)),
+     np.array([float.fromhex('-0x0.0000000000001p-1022'), float.fromhex('-0x1.74385446d71c3p+9')],
+              dtype=np.float64)),
 
     # Expected value computed using mpmath (with mpmath.mp.dps = 200) and then
     # converted to float.