[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
precision octave vs matlab
From: 
michaelschmid1 
Subject: 
precision octave vs matlab 
Date: 
Thu, 1 Jun 2006 07:03:50 +0200 
I don't think it's a octave bug. If it's a bug, I think I'm the culprit :(
The complete algorithm need several hundred lines of code... so I don't think
I will put this here...
I will do some more search and if I will find a position where I don't know
why a difference occurs... I will ask again...
but then with code ...
thanks
michel
On 29May2006, address@hidden wrote:
 Hi all

 First thanks for all the answers about my "global" problem... it works
now...

 but now the next problem..

 I have programmed some parts of a neural network toolbox for Octave, like
 the one of matlab. Some commands are working like "newff, train, sim" and
 so on..
 For the first training algorithm I decided to realize LevenbergMarquardt.
 Now if I make a comparison between the results of matlab and octave, there
 are some small differences...
 I searched for calculating errors but can't find anything. As far as I
know,
 both are using doubleprecision ... so where does this difference come
from?
 Is it really a calculating error or only a "rounding error"?

 Octave output:
 ================
 simOut =

 Columns 1 through 8:

 1.67883 2.05781 1.21596 2.82417 2.06661 2.00662 2.99825 1.65494

 Columns 9 through 16:

 1.65087 1.90819 1.90195 1.70142 2.61220 1.99405 0.99050 1.91911

 Columns 17 through 24:

 1.94686 0.93073 1.87499 1.75138 1.71708 2.21930 0.94781 2.02244

 Columns 25 through 30:

 2.01943 1.10781 2.02009 1.85863 2.71098 1.30415

 matlab output:
 ==============

 simOut =

 Columns 1 through 17

 1.6767 1.9494 1.1277 2.7196 2.0406 1.9435 2.8583
 1.6444 1.6437 1.8839 1.7769 1.6008 2.7412 1.9360
1.0998
 1.8426 1.8144

 Columns 18 through 30

 1.0464 1.8657 1.7506 1.6612 2.1693 1.0655 1.8792
 2.0094 1.0119 1.9947 1.8084 2.5923 1.2365

 Of course both programmes have the same initial conditions...

 I hope someone have similar experience with this kind of "trouble"
 and that this person can tell me that I didn't make a calculation error..
Given the information you have provided (just the outputs and no code)
how can anyone tell?
If you think you have found a bug in Octave, then please submit a
complete bug report to the address@hidden list. But please read
www.octave.org/bugs.html first to learn how to submit a useful bug
report that provides enough information to allow someone to actually
fix the problem.
Thanks,
jwe
[Prev in Thread] 
Current Thread 
[Next in Thread] 
 precision octave vs matlab,
michaelschmid1 <=