From: J. Wayde Allen (wallen@lug.boulder.co.us)
Date: 11/06/00-03:18:01 PM Z
On Mon, 6 Nov 2000, Jeffrey D. Mathias wrote:
> For a commertial site I still find it strange that no identification is
> given as to who these so called "product developers" are.
I'm not too sure what your beef is with this? Especially since you don't
seem to feel that the material presented is wrong. Most commercial sites
don't specifically list all of the people who worked on a product, but
rather just give contact info. What I found was:
To discuss how you can benefit from our consulting services, call Diane
at 203-790-8984 or send an e-mail to DianeA@pdlab.com.
Anyway, this is but one commercial site offering such services of
many. I don't really know anything about them. I just found them on the
web, and thought that they had a fairly well written description. Sorry
you didn't like it.
> In general and incomplete simplistic terms it is not wrong.
Yes, and that was why I posted it.
> But then it is the incomplete simplicity that can be misinformative.
Uh huh, but that is kind of what I'm getting at.
> No where are boundary conditions discussed (except hinted at as a
> "ranges in the variables"). Assessing boundary conditions is as
> important as selection of the variables.
Such things are only meaningful for a single experiment. One can't pick
variable ranges and boundary conditions without specifying the test
first. That doesn't invalidate the concept.
> Errors are oversimplified and no discussion is given to pitfalls of
> error accumulation, round-off errors, or limitations due to measuring
> devices. All seem to be invisibly simplified into the "bias error".
> And why not discriminate between linear and non-linear variables, or
> isolated and interdependent variables.
Again such things depend on the particular test done, and granted there
isn't a lot of such discussion on the web page I posted. At the same
time, I can't say I've ever seen much discussion of these things in the
testing results you've submitted to this list. Does that invalidate your
tests too? I certainly don't think so. Besides, everything you've listed
above is covered in the theories of multivariate testing. You can find it
all in the references I've provided. As far as the linear versus
non-linear variables goes, it is definitely a problem. Essentially none
of our processes have linear responses which makes the problem of variable
correlation even worse.
> There are times when changing one thing at a time will work, there are
> times when it won't. There are times when statistical analysis can
> provide insight, there are times when it is gibberish within a
> multitude of noise.
Very true.
> And what about the model, there has to be a statistical model. Part of
> the real problem comes when a statistical model is introduced. It seems
> more often than not too many assumptions, exaggerations, or
> political/financial decisions influence the results.
There is always a model whether explicit or implied. You have a method
you use for creating your platinum prints. I rather doubt you simply
throw a bunch of stuff together at random.
> As far as the art of alt-photo is concerned, perhaps results are
> achieved quicker by making prints through trial and error changing
> things based on experience and intuition.
Perhaps, yet at the same time there is a LOT of testing going on behind
the scenes here on the list. You've published quite a few tests yourself.
I was hoping I could possible offer a tiny bit of insight into a common
complaint, and also make people aware of some other methodology that could
possibly simplify this testing.
> So yes, I feel that site (pdlab) incompletely oversimplified what is
> necessary to perform an accurate statistical analysis.
But as I've noted before, the purpose , from my point of view anyway,
wasn't to teach the method, but simply explain the concept and make people
aware.
> And, I question if such an analysis is even the best route to solving
> most alt-photo optimizations.
You got a better way? If so you'd better write this up because there are
a lot of companies out there with BIG BUCKS that would sure like to hear
from you. Nevertheless, you are most certainly entitled to your own
opinion.
- Wayde
(wallen@lug.boulder.co.us)
This archive was generated by hypermail 2b30 : 12/01/00-11:46:56 AM Z CST