Modern math

We have a folder called PROTOS on a network drive. Every experiment gets assigned a controlled sub-folder, named something like J:\Protos\Z420_Joule_Tester, and there is a log file for them. We dump everything there: a readme file, whiteboard sketches, photos of everything, links, data sheets, test results, anything useful. The PROTOS folder of course gets backed up.

I take and graph a lot of data by hand, so I photograph that for the archive.

John Larkin wrote in news: snipped-for-privacy@4ax.com:

Bullshit.

You dissing the space program is 100% undeniable proof that you stopped learning decades ago.

John Larkin wrote in news: snipped-for-privacy@4ax.com:

I will likely outlive you by double. I am going to make at least

John Larkin wrote in news: snipped-for-privacy@4ax.com:

No, you lie. Because you piss and moan about things you are too goddamned lazy to understand or make use of and you even diss those whom do make use of them.

You have deliberately put on a pair of horse blinders to augment your tunnel vision.

And it is obvious, yet just like Trump, you do not even see it.

John Larkin wrote in news: snipped-for-privacy@4ax.com:

I refer to mine as "NPI" for New Product Introduction.

I prefer the formal aspects of my work to actually be formal.

"PROTOS" isn't even a word or an acronym.

We automate production testing, to minimize tech time. Some things take two hours to test and cal, but 5 minutes of that is tech time.

We take no production test data by hand (or by spreadsheet!) I do plot stuff myself, by hand, but that's engineering. Sometimes I plot it on my whiteboard and photograph that. I have a nice faint-grid whiteboard just to the right of my bench, so I can scribble easily while I'm testing.

The PROTOS archive is not for new product development. Once a product is defined, it gets its own working folder somewhere else.

J:\PROTOS is for experiments, good and failed.

It's a secret code word.

The issue that I raised is math precision. Once you zoom a bunch of times, single floats can't work. A few more, and doubles won't work.

John Larkin wrote in news: snipped-for-privacy@4ax.com:

So... no grasp of what an iterative process is then.

Here's some detail about the math precision problem:

Intel CPUs can do 10 byte reals although few compilers today support them in high level languages. Withdrawn after MSC v6.0 if memory serves for "compatibility" reasons. It was useful to have them available.

Once you go beyond a 64bit mantissa you have to roll your own.

Or use Rexx.

You can use NUMERIC DIGITS 10**6. Not the speediest, of course. ;)

Cheers

Phil Hobbs

Very OS/2 geek.

IBM FORTRAN had REAL*16 which was quite handy for moderately extreme precision numerical calculations at a still decent speed.

REAL*10 on the Intel kit was useful if you were summing a lot of vector product terms and didn't want the effort of sorting them small to large.

PowerBasic has an 80-bit float type. And 64 bit ints.

Divides must get interesting.

John Larkin wrote in news: snipped-for-privacy@4ax.com:

Well, it does it the schoolboy way, so it's a quadratic-time algorithm.

Simple stuff like 1/7 to a million digits is quicker than I can conveniently measure.

I wrote a little test program to test division of real N-digit numbers:

/* divtimer.rex */ arg digits . say "Computing (1/7)/(1/3) to" digits "digits" numeric digits round(digits) start = time("elapsed") say (1/7)/(1/3) end = time("elapsed") say "elapsed time" end-start "seconds"

The results are

digits time

10000 3.8 s

100000 393.25 s

So a full-precision million-digit divide would take about 11 hours on my laptop.

Cheers

Phil Hobbs

Fixed line 4.

Cheers

Phil Hobbs

Phil Hobbs wrote in news: snipped-for-privacy@supernews.com:

Not if you got your GPU to do it.