d/p/binomial-dist-median.patch: Update with upstream commit

1 files changed, 59 insertions, 15 deletions

diff --git a/debian/patches/binomial-dist-median.patch b/debian/patches/binomial-dist-median.patch
index 89477be..4636e64 100644
--- a/debian/patches/binomial-dist-median.patch
+++ b/debian/patches/binomial-dist-median.patch
@@ -1,22 +1,66 @@
-Description: Drop test for median value of the binomial distribution
- One of the BISTs in file BinomialDistribution.m hit a corner case.
- Indeed, the median value of a binomial distribution for n = 5 and p =
- 0.5 is not unique and the BIST produces different values on i386 and
- amd64 architectures (2 and 3, respectively). Hence, the package FTBFS
- on i386 without this patch.
-Author: Rafael Laboissière <rafael@debian.org>
-Bug: https://github.com/gnu-octave/statistics/issues/140
-Forwarded: not-needed
-Last-Update: 2024-03-10
-
---- octave-statistics-1.6.5.orig/inst/dist_obj/BinomialDistribution.m
-+++ octave-statistics-1.6.5/inst/dist_obj/BinomialDistribution.m
-@@ -662,7 +662,7 @@ endfunction
+Description: Fix median method of BinomialDistribution
+Author: Rafael Laboissière <rafael@laboissiere.net>
+Origin: upstream, https://github.com/gnu-octave/statistics/issues/142
+Last-Update: 2024-03-21
+
+From 8fdd588eb46a25e9ed9c3ab42286e44276632bf1 Mon Sep 17 00:00:00 2001
+From: Rafael Laboissière <rafael@laboissiere.net>
+Date: Tue, 19 Mar 2024 11:57:34 +0100
+Subject: [PATCH] Correct median value of binomial distribution when p = 0.5
+ and N is odd
+
+For a thorough discussion on this issue, see:
+
+Nowakowski S (2021) Uniqueness of a median of a binomial distribution with
+rational probability. Advances in Mathematics: Scientific Journal
+10(4):1951-1958. (http://dx.doi.org/10.37418/amsj.10.4.9)
+
+Essentially, when p = 0.5 and N is odd, there would be two "median
+values" (both integer values around N/2). In this case, the median value
+should be the fully trimmed mid-range, which happens to be equal to the
+mean value.
+
+Notice that these changes only cope with the untruncated case. For the
+general truncated case, the solution seems not to be straightforward.
+In particular, the code below:
+
+```
+pd = BinomialDistribution (5, 0.5);
+t = truncate (pd, 0, 5)
+t.median ()
+```
+
+will produce value `3`, which is wrong (it should be `2.5`).
+
+At any rate, this change is bug-compatible with Matlab, thanks to the
+use of `__traditional_()`.
+---
+ inst/dist_obj/BinomialDistribution.m | 8 ++++++--
+ 1 file changed, 6 insertions(+), 2 deletions(-)
+
+diff --git a/inst/dist_obj/BinomialDistribution.m b/inst/dist_obj/BinomialDistribution.m
+index c87e606b..3a1c5641 100644
+--- a/inst/dist_obj/BinomialDistribution.m
++++ b/inst/dist_obj/BinomialDistribution.m
+@@ -284,7 +284,11 @@ function disp (this)
+         Fa_b = binocdf ([lx, ux], this.N, this.p);
+         m = binoinv (sum (Fa_b) / 2, this.N, this.p);
+       else
+-        m = binoinv (0.5, this.N, this.p);
++        if (! __traditional__() & this.p == 0.5 & rem (this.N, 2) == 1)
++          m = this.mean ();
++        else
++          m = binoinv (0.5, this.N, this.p);
++        endif
+       endif
+     endfunction
+ 
+@@ -662,7 +666,7 @@ function checkparams (N, p)
  #%!assert (iqr (t), 1);
  %!assert (mean (pd), 2.5, 1e-10);
  #%!assert (mean (t), 2.8, 1e-10);
 -%!assert (median (pd), 3);
-+#%!assert (median (pd), 3);
++%!assert (median (pd), 2.5);
  #%!assert (median (t), 3);
  %!assert (pdf (pd, [0:5]), [0.0312, 0.1562, 0.3125, 0.3125, 0.1562, 0.0312], 1e-4);
  #%!assert (pdf (t, [0:5]), [0, 0, 0.4, 0.4, 0.2, 0], 1e-4);