home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
World of Ham Radio 1997
/
WOHR97_AmSoft_(1997-02-01).iso
/
antenna
/
ant_21
/
amsoft.iii
next >
Wrap
Text File
|
1997-02-01
|
4KB
|
87 lines
(version 1.21, 11/30/95)
Stat package put together from several textbook sources, but modified
considerably with many, many hours of work. Well, about 50 hours of work.
(Credit is given for the original source code: textbook sources.)
So far I have found that the Nth-order Regression program works the best for
fitting curves you are likely to find in e.g. the ARRL Antenna Book. You can
see from the graphs how close the program fits the curve: the curve with the
little triangles is the x,y data you input, and the curve without the marks
is the equation's fit using your x inputs. Caution: don't extrapolate using
the equation, if/when you incorporate the equation into your own program,
unless you test the extrapolation first. Otherwise you may get surprises.
The .com file mode43.com is included if you want to use ega43-line /
vga50-line mode, or switch back and forth between the two, without having
ansi.sys loaded as a device in your config.sys file. (If you have ansi.sys
loaded as a device, then the commands on the dos line are:
mode 80,50 to go to 50-line mode
mode 80,25 to go to 25-line mode.)
This package was designed to produce a somewhat pleasing output using 25-line
mode. You could re-write some of the source code to fit more on a screen,
recompile the programs, and then use mode43.com or the appropriate command
with ansi.sys.
But when the graphs are produced, and then the program ends, you get thrown
back into 25-line mode anyway.
The compiler used (11/94) was QuickBasic 3.0 for math co-processor, but with
emulator for non-math-coprocessor machines. A revision of bye.bas was
made 7/95 and compiled on QB 4.5, no co-processor. When i went to version
1.2 and 1.21, i've used the QB 4.5 no co-processor.
One final note, relating to a 'bug' i trapped and accounted for but did not
fully understand: at times, during execution of one of the regression
programs (such as Nth-order regression), you will find a listing like:
Standard error of estimate =
i.e., no value is given. When working with the source code and getting this
package together, I found that the original source code (out of the text-
book) at times was causing the program(s) to crash because there was division
by zero or square root of negative numbers. Rather than spend more time
reviewing my statistics and yet more time modifying the original textbook
algorithms, I simply put in an error trap giving the result displayed above
(no value given). Everything else, even in the cases where the 'blank'
appears for the Standard Error of Estimate, works just fine. The
correlation coefficient is correct, the graphs are correct, and the equations
produced by the program are correct; test it and see. Any suggestions on
this will be appreciated. The source code is here.
ANOTHER NOTE, on interpreting the screen with the graph on it: if you
see the equation
y = f(x) = (3.33334 * x ^ 1) + ( -4.3435 * x ^ 2) + -3.5
then you know that it would read something like:
2
y = f(x) = -4.3435x + 3.33334x - 3.5
in standard algebraic notation. My program's output gives the terms (the
powers of x) in ascending order, but standard notation for a polynomial
would be descending order. This goes almost without saying.
Orrin Winton
11/94, 7/95, 11/95
notes 11/30/95:
Discovered that geomreg and expreg, which take the LOG of input x,y
values, freak out when x=0 or y=0, because the log of (0) is undefined.
Today i set up some error trapping for this, and when the user inputs a
zero, the log is set to zero. This seems to work, and theoretically it
should work because as x approaches zero, the log of x also approaches
zero. Any comments from users out there?
Remember, i'm not a statistician. I got the original programs from
the textbooks i give credit to. They were supposed to know what they
were doing. So we gotta finish their work for them.