This is the mail archive of the gsl-discuss@sourceware.org mailing list for the GSL project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: preparing a new release


Make check gives me a lot of errors...

Let me know if I should paste more.
Cheers.
Alberto

FAIL:   SV_decomp_jacobi (5x5) case=33427394 [166953]
call returned status = 14
singular value 2 =    1.41421356237309492 vs previous    1.27608201443340596
FAIL:   SV_decomp_jacobi (5x5) case=33437404 [166963]
call returned status = 14
FAIL:   SV_decomp_jacobi (5x5) case=33438405 [166964]
call returned status = 14
FAIL:   SV_decomp_jacobi (5x5) case=33458425 [166984]
call returned status = 14
FAIL:   SV_decomp_jacobi (5x5) case=33489456 [167015]
FAIL: test
===================
1 of 1 tests failed
===================
make[2]: *** [check-TESTS] Error 1
make[2]: Leaving directory `/home/ambs/Junk/INCOMING/gsl-1.6.91/linalg'
make[1]: *** [check-am] Error 2
make[1]: Leaving directory `/home/ambs/Junk/INCOMING/gsl-1.6.91/linalg'
make: *** [check-recursive] Error 1
[ambs@eremita gsl-1.6.91]$


James Theiler wrote:
On Wed, 24 Aug 2005, Brian Gough wrote:

] I'm preparing release (it will be 1.7) -- I'd appreciate feedback
] on this test version:
] ] http://www.network-theory.co.uk/download/gsl/gsl-1.6.91.tar.gz
] http://www.network-theory.co.uk/download/gsl/gsl-1.6.91.tar.gz.sig
] ] If there are no problems reported then I will release it. Thanks.
] ]


configure; make; make check
all worked without a problem on RH Enterprise Linux 3

saw one typo in your list of changes:

** Fixed the branch selection in gsl_sf_gamma_inc_Q_e to prevent
caused inaccurate results for large a,x where x~=~a.

"prevent caused" -> "prevent"

===

By the way, I think there is an error in the documentation (code looks
okay) regarding Gaussian Tail Distribution, section 19.3.

Formula for p(x) should have an extra factor of 1/sqrt(2*pi*sigma^2).

Also, the figure for p(x) is incorrectly normalized -- you can
tell it is incorrect because the area under the curve should be one,
but is obviously much less than that.  It looks like that curve
was generated with a p(x) that *did* have the 1/sqrt(2*pi*sigma^2)
but did *not* have the 1/N(a;sigma) -- that is, it looks like the
tail of the gaussian in the Figure before it in section 19.2, without
any renormalization.

(An example of a correct picture of a tail distribution is in 19.10,
for the Rayleigh tail distribution.)


===


More on documentation, Chapter 7 introduces the Pochhammer symbol (a)_x with the function gsl_sf_poch. Four functions later, in
gsl_sf_pochrel, a new notation is introduced (a,x). If you want to
include (a,x) as an alternative notation, that could be done with
the gsl_sf_poch function; simpler would be to drop it entirely, and
say for gsl_sf_pochrel, ie:


These routines compute the relative Pochhammer symbol @math{((a,x) -
1)/x} where @math{(a,x) = (a)_x := \Gamma(a + x)/\Gamma(a)}.

->

These routines compute the relative Pochhammer symbol @math{((a)_x -
1)/x} where @math{(a)_x := \Gamma(a + x)/\Gamma(a)}.

===

The documentation in section 7.19 for gsl_sf_beta_inc does not define
B_x(a,b), which is
    B_x(a,b) = \int_0^x t^{a-1} (1-t)^{b-1} dt

===

On a more general note, I'd recommend changing the title of Section 7.19 from "Gamma Function" to "Gamma and Beta Functions".

And this is more for the ToDo list, but we could use the introduction to this section to talk about factorials, combinations, binomial coefficients, use of tables for integer values, etc.

As a quickfix, we might just remark on the identity \Gamma(n)=(n-1)!
for integer n>0.  I don't think this is in that section at all.


regards, jt


-- Alberto Simões - Departamento de Informática - Universidade do Minho Campus de Gualtar - 4710-057 Braga - Portugal


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]