Well, I've started modeling work on a project at work. So far it has been data gathering planning (what kind of data do we need) and asking lots of questions about the system I'm trying to model. And, while that is not over, now that I'm back in Pittsburgh, time to get started. Right now, I'm actually doing the "make the first one to throw away" thing. Because this is so amorphous, I'm not sure just how to solve this just yet anyway. I told the project manager this, and he actually likes the idea as a risk management principle. (because version 2 is supposed to be deployable, let's just forget about that until I have a model that works, then worry about making it deployable)
Now, a lot of this is the fact that I'm working in a domain that not too many modelers have gone in. Slides that I've built for presentations in the past have gotten real good reception, as people who have seen this get the idea, and the possible impact. Of course now I have to deliver.
Since I get to work on that most idealized of environments, clean slate and zero previous work to build on, I thought I would actually try to do things right. And take the opportunity to build up a new tool chest. For now, documentation is in LaTeX, with Dia to build up flow charts and UMLish diagrams. I'm reading Head First Design Patterns by Freeman and Freeman, and some ideas have gotten in my head.
For the model building, there is an idea in the back of my head that deployment will be inside MS Excel and Visual Basic (because everything quantitative is deployed inside MS Excel and Visual Basic). But because the tools are just hard to develop in, and it quickly turns unmanageable, even when done right, I'll probably use Python and/or R to develop the initial model. One big benefit is the various algorithm libraries that are readily available in Python and R as well as the automatic documentation tools like Docstrings, PyUnit and Sweave. And I've actually used all of them.
For the tool chain, LaTeX and Dia and Python. But, since I had to get a new laptop anyway, and needed to reinstall my tools, why not try something new. I'm doing almost everything inside of the Eclipse IDE, specifically the EasyEclipse version of it. In addition to Java development (which I may end up doing non of, unless I switch into Jython), it handles LaTeX (through TeXlipse), Python, through Pydev), database access, and version control through Subversion. First time I've used many of these tools before. It is rather nice. TeXlipse is much better then the JEdit or Vim that I've used in the past since it does command completion, and it seems even better then TeXnicCenter that I did my thesis in. Having version control as an integral part of the IDE is wonderful. I've tried it in the past and it just got too bulky, even with a graphic interface. Now it is just a button click. Auto-build in Eclipse is nice too. Having a window with the results updated with every save means near-realtime feedback. Since Dia saves diagrams as XML, that works in a version control system also. So all my notes, the documentation, diagrams and the source code will all be in version control. I almost feel like I know what I am doing with this development thing.
Of course, the real fun starts next. Going from all my diagrams to actual code and data.
Wednesday, October 03, 2007
Computer development platform
Labels:
Dia,
Eclipse IDE,
LaTeX,
programming,
Python,
R-Project,
subversion,
version control
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment