MAINT: stats: avoid spurious warnings in ncx2.pdf

scipy/stats/_distn_infrastructure.py

@@ -13,6 +13,7 @@
 from itertools import zip_longest
 
 from scipy._lib import doccer
+from scipy._lib._util import _lazywhere
 from ._distr_params import distcont, distdiscrete
 from scipy._lib._util import check_random_state
 from scipy._lib._util import _valarray as valarray
@@ -560,12 +561,22 @@ def _ncx2_log_pdf(x, df, nc):
     df2 = df/2.0 - 1.0
     xs, ns = np.sqrt(x), np.sqrt(nc)
     res = xlogy(df2/2.0, x/nc) - 0.5*(xs - ns)**2
-    res += np.log(ive(df2, xs*ns) / 2.0)
-    return res
+    corr = ive(df2, xs*ns) / 2.0
+    # Return res + np.log(corr) avoiding np.log(0)
+    return _lazywhere(
+        corr > 0,
+        (res, corr),
+        f=lambda r, c: r + np.log(c),
+        fillvalue=-np.inf)
 
 
 def _ncx2_pdf(x, df, nc):
-    return np.exp(_ncx2_log_pdf(x, df, nc))
+    # Copy of _ncx2_log_pdf avoiding np.log(0) when corr = 0
+    df2 = df/2.0 - 1.0
+    xs, ns = np.sqrt(x), np.sqrt(nc)
+    res = xlogy(df2/2.0, x/nc) - 0.5*(xs - ns)**2
+    corr = ive(df2, xs*ns) / 2.0
+    return np.exp(res) * corr
 
 
 def _ncx2_cdf(x, df, nc):
@@ -2519,7 +2530,7 @@ def expect(self, func=None, args=(), loc=0, scale=1, lb=None, ub=None,
         --------
 
         To understand the effect of the bounds of integration consider
-        
+
         >>> from scipy.stats import expon
         >>> expon(1).expect(lambda x: 1, lb=0.0, ub=2.0)
         0.6321205588285578

scipy/stats/tests/test_distributions.py

@@ -3739,13 +3739,19 @@ def test_ncx2_tails_ticket_955():
 def test_ncx2_tails_pdf():
     # ncx2.pdf does not return nans in extreme tails(example from gh-1577)
     # NB: this is to check that nan_to_num is not needed in ncx2.pdf
-    with suppress_warnings() as sup:
-        sup.filter(RuntimeWarning, "divide by zero encountered in log")
+    with warnings.catch_warnings():
+        warnings.simplefilter('error', RuntimeWarning)
         assert_equal(stats.ncx2.pdf(1, np.arange(340, 350), 2), 0)
         logval = stats.ncx2.logpdf(1, np.arange(340, 350), 2)
 
     assert_(np.isneginf(logval).all())
 
+    # Verify logpdf has extended precision when pdf underflows to 0
+    with warnings.catch_warnings():
+        warnings.simplefilter('error', RuntimeWarning)
+        assert_equal(stats.ncx2.pdf(10000, 3, 12), 0)
+        assert_allclose(stats.ncx2.logpdf(10000, 3, 12), -4662.444377524883)
+
 
 @pytest.mark.parametrize('method, expected', [
     ('cdf', np.array([2.497951336e-09, 3.437288941e-10])),