[parsec-users] Output of Blackscholes correct ?

Yungang Bao ybao at CS.Princeton.EDU
Wed May 25 15:27:59 EDT 2011


Hi Olivier,

Maybe you could try this link to download the file: http://parsec.cs.princeton.edu/download/2.1/parsec-2.1-sim.tar.gz

I tried blackscholes in 2.1. There are no errors for 1e-4 but a large number of errors for 1e-5. However, when I change the data type from "float" to "double", the errors disappeared for 1e-5. 

Yungang

----- Original Message -----
From: "Olivier Temam" <olivier.temam at inria.fr>
To: "PARSEC Users" <parsec-users at lists.cs.princeton.edu>
Sent: Wednesday, May 25, 2011 4:49:44 AM
Subject: Re: [parsec-users] Output of Blackscholes correct ?


Hi Yungang, 


On Tue, May 24, 2011 at 4:11 AM, Yungang Bao < ybao at cs.princeton.edu > wrote: 


Hi Olivier, 

The errors result from the float precision problem. You can change "fptype" (defined in blackscholes.c:42) from float to double to let blackscholes use double-precisioin. Then the errors can be eliminated. 



Thanks. I have tried and it did not change the error rate. But then, I forgot to mention I was using parsec-1.0. I downloaded parsec-2.1, tried again (with float, not double) on the small test input set and there is no error. 
I have been trying to confirm on large input sets since yesterday, but I failed so far to download the input sets (Simulation Inputs); about 20 download attempts so far, and all downloads stopping after a while (any other way to access them ?). I will keep trying, but then, it may be more that something was wrong in the 1.0 version. I noticed the error criterion has been reduced from 1e-5 in 1.0 to 1e-4 in 2.1, but the error was anyway much more significant in 1.0, so I don't believe that's the reason. 

Take care, 
Olivier 


P.S: Good to hear from you ! 



Best Regards, 
Yungang 




----- Original Message ----- 
From: "Olivier Temam" < olivier.temam at inria.fr > 
To: parsec-users at lists.cs.princeton.edu 
Sent: Monday, May 23, 2011 12:09:05 PM 
Subject: [parsec-users] Output of Blackscholes correct ? 


Hi, 


I am looking into the output of Blackscholes because I need to compare this algorithm against another algorithm performing similar option pricing predictions (and then compare their performance/implementation). 
In order to see how good the current implementation of Blackscholes was, I #define ERR_CHK and I was surprised with the very large number of errors (99.99%) as defined by the test, later on in the code ("fabs(priceDelta) >= 1e-5"). Assuming the current implementation is correct, what parameters should be changed in order to get a reasonable amount of errors (and thus decent outputs) ? 

Thanks in advance, 
Olivier 


P.S: The commands I used (I tried with various inputs): 

bin/parsecmgmt -a build -p blackscholes -c gcc-serial 
bin/parsecmgmt -a run -p blackscholes -i simsmall -c gcc-serial 


_______________________________________________ 
parsec-users mailing list 
parsec-users at lists.cs.princeton.edu 
https://lists.cs.princeton.edu/mailman/listinfo/parsec-users 
_______________________________________________ 
parsec-users mailing list 
parsec-users at lists.cs.princeton.edu 
https://lists.cs.princeton.edu/mailman/listinfo/parsec-users 


_______________________________________________
parsec-users mailing list
parsec-users at lists.cs.princeton.edu
https://lists.cs.princeton.edu/mailman/listinfo/parsec-users


More information about the parsec-users mailing list