wilmott.com

being looked down upon by "superior intellects" is one of privileges this forum freely grants,and as one wise wilmotter once said: "like prime numbers, stupidity is also Aleph(0)".cheers,Serp

Quote Gentlemen,Let's be gentlemen .Gentlemen, please no fighting in the War room as Cuch would say. This is beginning to look like the OT. We need a few pictures of buxomy ladies with chocolate smear on their pneumatic bits.

QuoteOriginally posted by: renormGentlemen,Let's be gentlemen .Outrun, I saw your post in boost mailing list. My understanding is that boost isn't optimal for specialized libs. What is good for general purpose use isn't necessary good for narrow domain use.Once our project gets rolling (hopefully very soon!) it is going to be unstable for quite some time. We can decide to boost or not to boost later.Personally I am interested in things that can't be done easily in Matlab, Python or even Java/C#. Nobody should be able to say "I can do it in C# 3 times faster" To be more specific, lets focus (at least at the beginning) on multi-factor Monte-Carlo and PDE solvers. How about 100 factor Bermudan LLM or portfolio with 100 or more assets? These types of simulations can completely bog you down unless carefully implemented in C++ or CUDA. I have some working LLM code that I can submit in very near future."No fighting the the War room." as Cuch pointed out.Renorm, you were the one suggesting layering the libraries and I agree with you. The first library to nail down should be the numerics level. Special function, random number generators, linear algebra. The Monte Carlo engine and PDE solvers seem to belong to the analytics layer to me. I hate to repeat myself, because I hate repeating myself, but before fingers can start tapping out code we need to settle on organizational issues. We need to pick a location that all developers have access to. I also think one key to a successful project is to have a project lead or co-leaders that can keep things moving in a forward direction.I've set up a survey for potential project sites. If anyone has suggestions for ones I've left out, please let me know and I will add them. FuseTalk has survey functionality, but evidently I am not a sufficiently superiour human being to figure out how to use it.

Here is my suggestion for a project (equity-oriented):A good set of Calibration tools that can either fit a real-world process to a time series or a risk-neutral process to an option chain. I do this in Mathematica, and am not a C/C++ programmer, so can't help out on that front. I would be willing to contribute some algorithms.One key to this type of project, IMO, is access to high quality data. I think youwould need a sponsoring company (like the Quantlib story) that was willing to contribute access to some good high frequency historical stock data and historical (end of day) option chains to make it worthwhile. In the end, there willbe licensing issues with the data, so that may be an insurmountable hurdle ..

kimosabe,All of those project hosting sites are fine. The final decision should belong to those who are willing to manage the project. Not that I am not willing to help to maintain the website, it is just I don't have much project hosting experience. I will just follow the lead.Alan,I can convert your Mathematica algorithms into C++. No working code from a knowledgeable researcher should be wasted. C++ knowledge shouldn't be a prerequisite. We could implement a feature that will allow submission of non C++ code and working papers explaining the code. Sourceforge users can simply upload zip file that everyone can download and review. Google code and Codeplex should support such feature too.Once we decide where to host our project, let's start implementing the numerical layer first. I will write a small working parer about parallel Monte-Carlo framework and submit it together with a tentative implementation.

And this is how it is done, gentlemen. Somebody steps up to the plate and takes a swiing.The two most difficult problems are who is going to bell the cat and what are we going to call the project. I''m happy with Cuch being the lead and am willing to do the menial chores of site maintainance. The survey stands at 3 for Google and 1 for Codeplex right now. Let's give it a couple of days to give people a chance to get familiar with the site capabilities.

You might want to consider github, too.

As Luigi pointed out, there are other attractive hosted repository solutions:- github- bitbucket- AssemblaMy understanding is that they all give free plans for open source projects. You need to think about svn versus distributed SCM, i.e. git or mercurial. Next, you want to consider the dev process. Anybody can just throw in code vs sandbox with reviews before code can be allowed in main repository a la Boost.Then Testing frameworks. Options for Unit test here are:- GoogleTest- Boost.Test combined with Turtle for MockingAt some point, you need to look into Build systems

QuoteOriginally posted by: outrunwhat about license?* e.g. boost is saying that the GPL and LGPL are too restrictive, and don't accept code under that license, they probably gave that some good thoughts* we should also see *if* we can pull in and refactor code parts of other project legally to speed up the processWe can choose from here: Open Source LicensesI vote for a license which allows commercial use.Dominik

QuoteOriginally posted by: outrun(...)* e.g. boost is saying that the GPL and LGPL are too restrictive, and don't accept code under that license, they probably gave that some good thoughts (...)The Boost Software License and explanationsDominik

QuoteOriginally posted by: madmaxAs Luigi pointed out, there are other attractive hosted repository solutions:- github- bitbucket- AssemblaMy understanding is that they all give free plans for open source projects. You need to think about svn versus distributed SCM, i.e. git or mercurial. Next, you want to consider the dev process. Anybody can just throw in code vs sandbox with reviews before code can be allowed in main repository a la Boost.Then Testing frameworks. Options for Unit test here are:- GoogleTest- Boost.Test combined with Turtle for MockingAt some point, you need to look into Build systemsDone, done, and done. Added to SurveyMonkey that now wants to survey me. What kind of soda pop do I like to drink? Maybe we should just hash it out here and pick one.I've gone the rcs + goofy scripts to make it usable, to cvs, then svn route. I'm not happy with svn. Why do I have to do cleanup when things stop working? I've started using mercurial and like it much better, but if git is good enough for Linus, I'm open to that too. I think distributed SCM is the way to go with a project like this. People can (and probably will) go their separate ways, but it seems to make merging more managable when people kiss and make up.Testing is essential, but I think we should keep it simple. We are only writing functions and lightweight classes.