wilmott.com

Doofus, ;-)I am quite interested in that paper, but I can't access the file. My curiousity is especially piqued when they compare the performance of the latest Java compilers to gcc. Do they use high performance compilers for the compiled code also? Using Compaq or Intel compilers (or any other compiler built for high performance) will make quite a difference.If anyone can read that as a .ps & email it to me, I would be grateful. I can't get winzip, gzip or winace to decompress it.matt.

Matt, Forwarde to your hotmail account , for anyone else interested here it is , if there is trouble with the zip file I will send to you as an email attachment

QuoteOriginally posted by: AnthonyI use all three in varying degrees. Matlab is very good for writing fast 'disposable' code to look at a problem - similar to VBA, doesn't take too long to get some results although plotting graphs exactly how you want them usually involves a trip through the help files. C++ is v fast and good for permanent code, dll /xll etc. I find if I dont use it all the time, I'm unlikely to get it to compile the first time round before it highlights a 'gap' in my knowledge on pointers etc, which is not very useful when you have 1/2 an hour to price a structure. Mathematica I'd like to use more, the symbolic features are good for checking algebra.At risk of opening this one up , although the thread seems to have been taken over by Java fiends, what is thought of other packages for numerical computation for rapid development deployment of financial pricing applications & time series analysis? (I am familiar with the MATLAB v Mathematica - numerical v symbolic process - arguments):Maple,MathCadSciLabNumerical Algorithms Group (NAG) LibrariesOctaveO- MatrixRogue Wave LibrariesR-QuantThese may be remembered fondly/ not so fondly from degree days...RATS, TSP, S-Plus, SPSSGaussI would especially appreciate any pointers to a decent graphical plotting package that could be driven from Excel VBA on a real time basis rather than via flat files that could be used (in DLL form?) to plot complex surfaces etc.. and even provide second level data landscape facilities.Does anyone remember a company that in the early/mid 90's were promoting data landscapes on Silicon Graphics workstations?. Some of the people invelved were psychologists who realised that data could be more easily absorbed in graphical, rather than numerical form & I think I heard they teamed up with some ex- military Head Up Display designers to produce a concept that data output could be summarised in graphical form in a GUI and then zoomed into to provide details on certain components in the hierarchy, or navigated through like a virtual 3d landscape. I think the demo they used was a model trading floor with numerous desks showing in various colours representing there positions & this could be expanded to show traders, products, currencies etc..I know the above is not strictly finance, but sometimes getting the best numerical output isn't great if the dataset is too large to be able to make sense of the output quickly.Thanks

There was a book called 'Free to Trade' by Michael Ridpath that touched on the subject of head sets in trading - i dont recommend the book though!In terms of surfaces in Excel, i've started using Excel Link by Matlab. You can write a function in Matlab, compile it into C code as a dll and then attach it to your worksheet. The function call can produce a Matlab surface from the Excel data within Excel. Simple and fast.

Anthony,Thanks for your answer I will definately try it out.

Matt,Thanks for your competent comment and reference to experts (I do recommend reading it by everyone doing FP in Java). Isn`t it suprising, that Intel processors, obviously addressing general markets, have FPU architecture outsmarting many other, apparently more professional, designs?

mrm,I remember TSP and SPSS from back when i was in school. TSP requires you to have a very good understanding of the underlying process so that you can write the code properly. SPSS was simply, enter data and choose your process from the menu. Simple, and the only thing you needed to know was how to interpret the output. From my experience SAS is more commonly used in practice. Basically it is similar to the feature in SPSS where you enter code.

mrm,From my own experience & knowledge:Maple-- good at symbolic math, available in Matlab symbolic toolboxMathCad -- no experienceSciLab -- similar to Matlab, but OSS. Less refined than commerical packages, especially plotting support, but the price is right.Numerical Algorithms Group (NAG) Libraries -- very good, but you must have a language to wrap them in (C/C++/Fortran/Java)Octave -- similar to SciLabO- Matrix -- Matlab wannabe. Very nice package, but its primary selling point was a JIT compiler for code, so that its speeds were better than Matlab. Matlab now has a JIT, too.Rogue Wave Libraries -- never heard of 'emR-Quant -- dittoRATS, TSP -- great time series analysis apps, but of questionable use for pricing. S-Plus, SPSS -- very good general purpose statistical apps, SPSS is more menu-driven, S+ is more script-oriented. I am unsure of usefulness for pricingGauss -- similar to Matlab for econometricians. For example, if you want to solve an MLE function in Matlab, you write a likelihood fn, and then minimize its negative, then you write or find code to compute the VCV and/or residuals or Langrangians on constraints, etc. Gauss has a maximum likelihood routine that does all that. It also has a Constrained Maximum Likelihood routine. I haven't used it in a few years, but I was unimpressed with it when I did. It just wasn't 'polished'. Matt.

mmatloch,I wouldn't necessarily call me competent on architecture by any means. I've just run into problems & had to read enough to solve them. As for the x86 FP architecture, I (and many others, including its creator) would hardly say that it outsmarts other designs. Its original designer, whose name I've forgotten, called the x87 FP architecture "brain-dead". It was a kludge then that has had many kludges piled upon it to try and keep up. The poor FP design of the x86 was one of the driving forces behind the switch to IA64, with a completely new instruction set, instead of AMD's choice of x86-64. The Alpha had a good FP architecture, the current IBM Power4 has a good architecture. The x86 has a crappy architecture, but some of the brightest people in the world have been paid very large amounts of money to kludge together enough temporary fixes to make it deliver reasonable performance in spite of itself.Matt.