Re: [boost] Help needed with some Math lib failures

Subject: Re: [boost] Help needed with some Math lib failures
From: Sebastian Redl (sebastian.redl_at_[hidden])
Date: 2009-03-09 15:08:33

John Maddock wrote:
> Folks,
>
> There are a few failures from Boost.Math (for example
> http://tinyurl.com/clvlq3) which I'm unable to reproduce locally - it
> seems only certain (possibly AMD x86-64 machines running Linux or
> *BSD) exhibit the problem.  I can provide a test case to help with
> debugging what's going on, but I need to find someone who can
> reproduce the issue first.  So... if you think you can help, then cd
> into libs/math/test and do a:
>
> bjam gcc test_hypergeometric_dist0 test_hypergeometric_dist1
> test_hypergeometric_dist2 test_hypergeometric_dist3
> test_hypergeometric_dist4 test_hypergeometric_dist5
>
> and if one or more fail then let me know the failure message and I'll
> put together a test case.
I can reproduce the failure for dist1, dist2, dist3 and dist5. The full
error message is too long for this list to accept, so I'm just posting
the message of dist1.

Linux x86-64, GCC 4.3.3

Running 1 test
case...                                                                                                                                      

Checking quantiles with Mathematica data with type
float                                                                                                    

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                          

Checking quantiles with Random large data with type
float                                                                                                   

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                          

Absolute
tolerance:1.19209e-07                                                                                                                              

Tolerance for type f is 0.000596046
%                                                                                                                       

Checking quantiles with Mathematica data with type
double                                                                                                   

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                          

Checking quantiles with Random large data with type
double                                                                                                  

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                          

test_hypergeometric_dist.cpp(240): error in "test_main_caller( argc,
argv )": check quantile(du, cp) >= x
failed                                            
Failure was with data x = 30520, r = 130605, n = 130605, N = 523360, p =
1.315394088731268169855067478460234e-53, q =
1                                     
Absolute
tolerance:2.22045e-16                                                                                                                              

Tolerance for type d is 1.11022e-12
%                                                                                                                       

Checking quantiles with Mathematica data with type long
double                                                                                              

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                          

Checking quantiles with Random large data with type long
double                                                                                             

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                          

test_hypergeometric_dist.cpp(240): error in "test_main_caller( argc,
argv )": check quantile(du, cp) >= x
failed                                            
Failure was with data x = 30520, r = 130605, n = 130605, N = 523360, p =
1.3153940887312682609778382091540272e-53, q =
1                                    
test_hypergeometric_dist.cpp(240): error in "test_main_caller( argc,
argv )": check quantile(du, cp) >= x
failed                                            
Failure was with data x = 4770, r = 130605, n = 130605, N = 1030167, p =
1.22798082982754627280320578058046e-3094, q =
1                                    
test_hypergeometric_dist.cpp(240): error in "test_main_caller( argc,
argv )": check quantile(du, cp) >= x
failed                                            
Failure was with data x = 15134, r = 256574, n = 130605, N = 1030167, p
= 1.1105244685326902267944526554403882e-3545, q =
1                                 
test_hypergeometric_dist.cpp(240): error in "test_main_caller( argc,
argv )": check quantile(du, cp) >= x
failed                                            
Failure was with data x = 11903, r = 256574, n = 256574, N = 2427690, p
= 1.9101122505923986270675426093883161e-2808, q =
1                                 
test_hypergeometric_dist.cpp(240): error in "test_main_caller( argc,
argv )": check quantile(du, cp) >= x
failed                                            
Failure was with data x = 241837, r = 523360, n = 523360, N = 1030167, p
= 2.6242122183575062859028854854238337e-1957, q =
1                                
test_hypergeometric_dist.cpp(240): error in "test_main_caller( argc,
argv )": check quantile(du, cp) >= x
failed                                            
Failure was with data x = 208651, r = 2063277, n = 256574, N = 2427690,
p = 4.0582793494211428626917554262141244e-625, q =
1                                
Absolute
tolerance:2e-16                                                                                                                                    

Tolerance for type e is 1e-12
%                                                                                                                             

Checking quantiles with Mathematica data with type
real_concept                                                                                             

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                          

Absolute
tolerance:2e-16                                                                                                                                    

Tolerance for type N5boost4math8concepts12real_conceptE is 1e-12
%                                                                                          

*** 7 failures detected in test suite "Test Program"

EXIT STATUS: 201