Re: tisean-0.2.3 : test failures

[Piotr Held]

2015-09-04 15:21 GMT-06:00 Marco Atzeri <address@hidden>:

On 04/09/2015 18:56, Oliver Heimlich wrote:

On 04.09.2015 18:37, Marco Atzeri wrote:

On cygwin 64 bit, two tests are producing notable
errors. All the others are passed.
Could someone confirm on other platform ?

Marco, all integrated tests pass on my Debian 64 bit system. Please let
me know if you need further details for identification of the error's
cause.

something looks wrong in this ikeda function

this should produce same of ikeda(1)

a          = 0.4;
b          = 6.0;
c          = 0.9;
x0         = 0.68587;
y0         = 0.65876;
TH=a-b/(1+x0^2+y0^2);
1+c*(x0+y0*j)*(cos(TH)+j*sin(TH))

ans =  0.65523 - 0.78338i

However on cygwin 64

> ikeda(1)
ans =

   1.21651   0.41790

on cygwin 32 bit

 ikeda(1)
ans =

   1.06022  -0.13853


and the expected value is

   0.28572    0.6434

I am a bit puzzled

Hi Marco,

OK, so ikeda is kind of weird. 

First of all please note that ikeda(1) gives you the 10001th element. 

A lot of things can happen when calculating the 10000 first elements. I don't know if this is the case here, but I found that '$ ikeda -l1' - the original TISEAN implementation would give:

 1.18407357      -1.33515096 

Now I think the TISEAN implementation uses floats instead of doubles.

Also the __ikeda__.cc is used only for optimization (it's a recursive algorithm, so I couldn't vectorize it). 

I don't know if that helps any...

Regards,

Piotr