[From Bill Powers (940202.0915 MST)]
Bill Leach (940201.1846) --
RE: programming.
One reason I prefer the PC environment is that our experiments
and simulations tend to use the whole capacity of the CPU -- even
the 33 MHz 486 DX machine I now use. In tracking and other
control experiments, I like to sample the handle position (the
subject's behavior) at least 30 times per second and preferably
60 or more, and this cycle time has to include erasing and
regenerating the display in real time, as well as all
computations that are used to relate the handle action to what
happens on the screen. The display and sampling rates have to be
synchronized with the frame rate of the screen. On multitasking
systems like the the Macintosh, the Amiga, or a Unix system, it's
pretty hard to make sure that the samples and display refreshes
actually occur at uniform intervals of real time synchronized
with the display rate. I had an Amiga for a while, and discovered
that the sampling rate depended on whether the mouse was moving,
and on how fast it moved -- all those interrupts and the enormous
overhead of multitasking screwed up quantitative data-taking. I
sold it and went back to PCs. On a Unix machine where you're
sharing the CPU with other users, I should think this problem
would be even worse. It's even bad on a Macintosh with only one
user.
Also, I'm in the same boat as you: I'm not an advanced
programmer, so the friendly environment of Borland C 2.0 is
perfect for me. Compile and run, or pop back into the editor with
the cursor on the error. I can't imagine going back to the
command-line mode. What is that GNU C program like to use? Do you
have to use a separate editor? I suppose I should try it before I
get others involved too deeply in Turbo C. Martin Taylor: is that
offer still good?
I'm avoiding going into OOPS, because it's really not structured
right for control theory. Associating objects with operations
means that a given control system, which is a series of
operations on variables, gets tied to specific objects in the
environment instead of being applicable to any objects it senses
and acts upon. So if a hand is used to lift a glass, that's a
different "object" from the same hand lifting a pencil or a book.
I looked into OOPS with Turbo Pascal 5.5, but didn't get turned
on by it. Don't really know a lot about it, though. Also, the
object code from C++, I have heard from authorities, is
considerably slower than vanilla C.
I tried Forth, too, and it is fun to write but hard to keep track
of (at least the way I write it). I think it appeals most to
natural mathematicians who have no difficulty remembering what
complex operations symbols stand for. That's not me. Also, Forth
was really optimized for a PDP11 -- it's not all that fast on a
PC.
I think that if we write our programs in Turbo C 2.0, we should
be compatible with ANSI machines except for some library
routines, like the graphics, and DOS-specific commands like
interrupts, sound, time functions, etc. Come to think of it, the
list grows -- it could be a pain.
Do you want me to send you some source code to see how hard it is
to port to your machine? The surest way to get _really_ hooked on
PCT is to do some experiments and see how well the models fit the
data!
BTW, when I don't answer some comments, it's not necessarily
because I don't find them interesting. I often can't think of
anything to add that's worth saying. As Dag says, "No comment, no
error" (hardly).
ยทยทยท
---------------------------------------------------------------
Mark Olson (940201.1752) --
Mark, you're talking about model-based control, an elaboration of
controlling imagined perceptions. This may be appropriate for
some of the higher levels of control. The problem at lower levels
is updating the model rapidly enough to handle real-time
disturbances (which the model can't anticipate). I haven't yet
seen any demos of this type of control, and haven't tried any
myself.
---------------------------------------------------------------
Best to all,
Bill P.
