Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 0 down vote favorite

I've been tasked with taking up some modeling (perhaps predictive) for chemical formulations and some resultant performance characteristics. I have experience with finite element analysis (mechanical and fluid) and some more basic statistics, which identified me as a candidate for taking this on. I'd like to learn more in order to better understand this request.

I'm being trained by someone and it seems the software my company is using for this is Dassault iSight.

My questions:

  • I'm a bit disadvantaged as it was hard for me to search this site without even understanding the proper terminology. I'm hoping someone with iSight experience (or from scanning the page) could let me know what "class" of software this is -- neural network? Machine learning? Bayesian modeling? Predictive analysis?
  • Might I be pointed to some resources to learn more about this type of modeling so I can understand what iSight is doing on a more theoretical level?
  • If someone knows iSight, are there other packages that aim to do the same? I'm not sure how they chose this software package; if something else would be better suited (preferably in R, which I have experience with), I'd love to know about it.

For reference and without getting very specific we've conducted a DOE of various chemical formulation ratios for a material and have five or so responses we've tested on each of our ~80 lots. Now we're looking to determine the interactions between the X's (formulation components) as well as how the X's affect the Y's (performance attributes).

I realize this is a total noob/ignorance question... just looking to understand the proper terminology and direction for learning more as it's been hard to search on my own.

share|improve this question
This doesn't appear to have anything to do with statistics. It looks to me like it is a suite of software products that provide an environment for you to do networking , simulation etc. But you would probably have to learn how to program the tools to perform your particualr application. I can't tell much from the description as to what it does to make it advantageous for you to have it. it might enable you to interface several simulated functions into one large simulation. – Michael Chernick Jun 8 '12 at 21:03
Was my summary of what we're trying to do helpful? As stated, we're using the software to help figure out the relationship between multiple X's as well as how they interplay to affect Y's. Knowing that, are there multiple ways to do this, or perhaps an umbrella term for how that sort of problem is approached? Is this simply multi-variable regression analysis of some sort? – Hendy Jun 8 '12 at 21:43
@Hendy, there are many, many ways to do this. Regression would be one sensible way to go, particularly if you have have some insight into plausible relationships between the Xs and Ys. At the other extreme, a classification/machine learning approach might help you predict new X->Y relationships even if you don't know much about how Xs influence Y. It seems like your software can do both. – Matt Krause Jun 8 '12 at 22:23
The site has no hint of any statistical methods. – Michael Chernick Jun 8 '12 at 23:24
@MichaelChernick Thanks for checking. There are several modeling techniques listed in the program: RBF, Kriging, Response Surface, and Orthogonal Polynomial. I'm going to look further into these and some of the other suggestions made. Thanks! – Hendy Jun 12 '12 at 16:23

1 Answer

up vote 3 down vote

I've never used iSight, so I took a quick look at the product description (pdf). It seems like a fancy data analysis package that has an (allegedly) easy to use, LabView-like GUI and a proprietary distributed computing engine.

The last two pages of the data sheet suggest that essentially does a lot of standard statistical stuff (e.g., generating experimental design matrices, fitting distributions, a smattering of machine learning) along with some numerical optimization, visualization, etc. I haven't checked every single thing in that pamphlet, but I'm pretty sure R/NumPy/Matlab can do everything on there, even if the workflow ends up being a little different.

If you've got questions about what various iSight components do (e.g., "What is this Latin Hypercube thing?"), they'd definitely be germane here. Questions about the statistical aspects of using the program are probably on-topic too, but you might have better luck with the network/distributed computing part on stackoverflow or superuser.

share|improve this answer
Thanks for this. I got some more training and the models provided for analysis are Radial Basis Function, Kriging, Response Surface, and Orthogonal Polynomial. I'll be looking more into these. My goal is to understand how these approximations work on a small scale to at least not have this remain a black box to me. In the process, I'd like to explore other solutions, primarily R based, just to see if there's more out there that could be useful. – Hendy Jun 12 '12 at 16:22
1  
Glad to help! As you may already know, CRAN has an astonishing number of R packages, which might give you some working code and pointers to references. Someone else has been asking about kriging, so you might want to take a look at those questions/answers too. – Matt Krause Jun 12 '12 at 20:43

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.