Unknown HP Classic Prototype

[aj04062]

Does anyone know what model this prototype front was for?

See image on my website:

[rprosperi]

Looks like an unreleased statistics-focused model in the Classic series.

Interesting find! No guts to go with it I assume??

[aj04062]

No guts.

[AnnoyedOne]

Ok, I admit it I had to look up ANOVA. I'd never heard of it.

https://en.wikipedia.org/wiki/Analysis_of_variance

A1

[dm319]

Wow, this is quite unusual! Haven't looked into it too much, but I imagine statistics would have been hard to do on the earlier machines as many statistical functions need to be calculated on the entire dataset rather than cumulatively. Which would have needed a lot of memory I guess. Having said that I don't know how linear regression, standard deviation etc are calculated on devices with limited memory.

[Thomas Klemm]

(10-27-2024 11:34 AM)dm319 Wrote:  Having said that I don't know how linear regression, standard deviation etc are calculated on devices with limited memory.

E.g. from HP-11C Owner's Handbook and Problem-Solving Guide pp 60.

Mean

\(
\bar{x} = \frac{\sum x}{n}
\)

\(
\bar{y} = \frac{\sum y}{n}
\)

Standard Deviation

\(
s_x = \sqrt{\frac{n \sum x^2 - (\sum x)^2}{n(n-1)}}
\)

\(
s_y = \sqrt{\frac{n \sum y^2 - (\sum y)^2}{n(n-1)}}
\)

Linear Regression

\(
A = \frac{n \sum xy - \sum x \sum y}{n \sum x^2 - (\sum x)^2}
\)

\(
B = \frac{\sum y \sum x^2 - \sum x \sum xy}{n \sum x^2 - (\sum x)^2}
\)

[naddy]

(10-27-2024 11:34 AM)dm319 Wrote:  Having said that I don't know how linear regression, standard deviation etc are calculated on devices with limited memory.

Σ+ accumulates n, Σx, Σx², Σy, Σy², Σxy in registers. It follows that the available functions can be calculated from those values. Of course Thomas already provided the formulas.

The HP-42S additionally accumulates Σln x, Σ(ln x)², Σln y, Σ(ln y)², Σln x ln y, Σx ln y, Σy ln x, from which it can also provide exponential, logarithmic, and power curve fits. The HP-11C Owner's Handbook and Problem-Solving Guide has a curve-fitting program (p. 162ff), which builds the same functionality on top of the L.R. function by using the logarithm or exponential of various values, so there the curve fit must be chosen before data entry.

[dm319]

(10-27-2024 02:34 PM)naddy Wrote:  Σ+ accumulates n, Σx, Σx², Σy, Σy², Σxy in registers. It follows that the available functions can be calculated from those values. Of course Thomas already provided the formulas.

Ah that makes sense. For ANOVA, from memory, you are fitting normal distributions to each group and then seeing if one group is outwith the population overall. But it has been a while since I've done an ANOVE in anger so I may have misremembered. Either way, it sounds like it might be tricky to find a sunmary statistic that you can accumulate to calculate the result later. Many of the more advanced statistical functions need access to all the individual data points, which would have been a challenge back then.

[John Garza (3665)]

I also see normal distribution, t-test, and chi-squared on the keyboard.

-J

[Harald]

OK, so now we need to get @teenix to write code for this and bring it to life

[teenix]

(10-29-2024 03:45 PM)Harald Wrote:  OK, so now we need to get @teenix to write code for this and bring it to life

I'm like the prototype on this one - no guts :-)

[Duane Hess]

Don't know squat about statistics. 1st reaction was possibly a handheld 9805 prototype?

But the functions don't match the 9805.

However, in terms of available/affordable memory at the time, the following imply its realistic to put ANOVA in a handheld:

https://www.hpmuseum.org/forum/printthre...?tid=12152

Chi square, t-test, ANOVA, normal distribution, if I understand correctly, is in the 67/97 Stat Pak-I. The point? Getting around to my assumption why a model like this never was produced.

So memory requirements for the unknown calc? The data registers of for a 67/97. Being a cardless, unlike the 67/97, all keyboard functions would be in ROM. Assuming the functions would be microcoded, would it be reasonable the memory needed would be less than the sum of the all the needed program steps from the Stat Pak?

Let's assume the unknown calc was a real HP project, not a 3rd party product but doing a knock off based on appearance. Also unknown(?) is what timeframe the prototype was being considered. The 35 cable out in '72 and the 65 only 2 years later. Unknown (to me) is how long was HP considering a machine like the 65.

The sum of the blah-blah is needed program memory would make a pricey product. Seems reasonable the project was dropped particularly if there was any glimmer in someone's mind a product like the 65 would be developed. Expensive due to card reader, but extensible and flexible due to the card reader. So why make a high cost fixed function prototype when a more flexible product is coming. Even though the 65 would be more $.

Over my head, relative to the above, is how much memory did the 45 & 80 need? Less or more than the prototype? Considering how fast technology was changing, i.e. the much lower cost Woodstock series.... Sounds like a feasible reason to drop a classic technology prototype.

Is it reasonable, considering the price/feature evolution of the 65, 67, 97 and 29C, 19C why bother? Or is a more reasonable viewpoint simply not enough of a market for a fixed function device? Be it classic technology or lower-cost Woodstock electronics.

Considering how little I care about statistics, not sure my post is needed. Must be hyped up on coffee. I certainly have plenty of that.

[Geoff Quickfall]

Must be hyped up on coffee. I certainly have plenty of that.

Statistically speaking coffee drinkers type faster but with more mistakes :-)

Used my HP41cx with stat pack for my MSc thesis back in 1986.

Worked well!

Now the Monroe / Compuserve had dedicated machines such as the 324, 326, 340 and the 354. The statistician was the 340 so I can see HP eyeing the competition in the 1970’s.

Cheers

To see a compucorp 326 and 392 in use:


Compucorp Scientist

[John Garza (3665)]

Let's not forget the huge Commodore pre-programmed machines, and I believe there was a Novus/NS Statistician model too.

[John Garza (3665)]

Weird thing is, there is absolutely no need for pre-programmed machines if you have a decent programmable. My guess is it was a marketing/price-point issue. Or maybe a slight edge on execution speed. On less technical products, one also considers 'dumbing it down' for the typical user and making it more foolproof. But I don't think that's the case with calculators of this type.

[aj04062]

(10-29-2024 03:45 PM)Harald Wrote:  OK, so now we need to get @teenix to write code for this and bring it to life

That would be pretty wild. I have guts from classics to make that happen!