Simple Lineal Regression in RPL & other issues
in [HP48]
|
Prev: Newer ROMs for HP 50g?
Next: System Fonts
From: lacro on 15 Jun 2008 23:00 Dear Leon, I think I need "more help with the code" and I don't discard your "few other programs to solve diferent things ... like obly (?), b0 and b1... OK! despite this very one (although very thorough), it doesn't fulfil the issue of getting all those outputs WITH ONLY the inputs requested (the accumulated sums), AND WITHOUT the data points as inputs. However, I've understood the 4th, 3rd and believe the *1st level stack concepts, but what did you mean in the 2nd stack by "Number of X..."? The "N" quantity of values, OR the amount of significant figures before the comma as fraction mark? I think in the last (1st) you specified the amount of figures after the comma. In any case, when I copy/paste the whole program from this very newsgroup to my HPUserEdit 5 application, it was thoroughly understood ... Although somehow, when I sent it through my Connectivity Kit , it changed some symbols in the way (like e.g. at the begining, are there 24 zeros in a row and more variables over that same number?) When there is a " LIST", I assume it means "S*LIST" (where S* stands for Summation Sign) ? Fortunately, the @ signs didn't show up (and there is no need to "clean" them then), and the TAG command didn't show the left arrow like in "->TAG"... I tried an example without success, A) Your Application, 4 { 5.5 7.2 9.4 10 12.2 } 3 { 2 2.5 3 3.5 4 } 2: 5 (data points) OR 1 for one figure before the comma ? 1: 1 (only 1 decimal after the comma) ? vs. B) My intended app, S*X = 15 S*Y = 44.3 S*X2 = 47.5 S*Y2 = 419.29 S*XY = 141 S*N = 5 with S* as Epsilon Sum SIgn OUTPUTS = almost as extensive as yours (there are many I won't get, because of the lack of information I would had get from the data points in your application INPUTS). Cheers, Carlos On Jun 14, 8:11 pm, Leon <leon...(a)gmail.com> wrote:
From: Leon on 16 Jun 2008 01:31 Hello Carlos obly suposed to be only now 4 { 5.5 7.2 9.4 10 12.2 } @ Y { valores de Y en lista } 3 { 2 2.5 3 3.5 4 } @ X { valores de X en lista } 2 5 @ n ( valor de n como numero no lista ) 1 1 @ number of decimals of Y if it has any decimales que tenga Y i will try to save a file with the code from HPUserEdit 5 and send it to you this week with the improved code
From: lacro on 16 Jun 2008 17:17 Fine with all. I understood you correctly then. It must be I didn't get the whole code of the application properly during the "translations" then. Tks anyway. Meanwhile and for the rest, the other challenges stay on. A proper way of computing the regression OUTPUTS, provided ONLY the 6 initial summary statistics (x, y, x^2, y^2, x*y, n) as the local variables defined (BUT WITHOUT the SDAT matrix values of whichever the number of actual data points). Alternatives to solve statisitical distributions like e.g. LN, Gamma, Weibull, Pascal, t, Chi2 and F-St at least (let alone the Normal / Std, Binomial and Poisson), along the way Stuart Sidders did with his 48SX release of STATSPACK49 (converted and ported by Gaak later), and perhaps even a way to get the distribution parameters computed in the way or to solve Partial Mathematic Expectation (for Mean) cases. A way to linearize expressions like 'Y = alfa * X^beta' in order to get 'ln Y=ln alfa + beta * ln X'... is it with the LIN function? A reasonable 4x series emulator for DOS (HP 200LX) along its ROM, or if somebody is so kind to explain me how to use the existing ones. Regards, Carlos ---------------------------------------------------------------- On Jun 16, 2:31 am, Leon <leon...(a)gmail.com> wrote: > Hello Carlos > obly suposed to be only > now > 4 { 5.5 7.2 9.4 10 12.2 } @ Y { valores de Y en lista } > 3 { 2 2.5 3 3.5 4 } @ X { valores de X en lista } > 2 5 @ n ( valor de n como numero no lista ) > 1 1 @ number of decimals of Y if it has any decimales que tenga Y > > i will try to save a file with the code from HPUserEdit 5 and send it > to you this week with the improved code
From: John H Meyers on 16 Jun 2008 19:06 On Sat, 14 Jun 2008 16:08:56 -0500, lacro wrote: > I couldn't find that very interesting distintion reference you mentioned. I didn't provide a reference (for the fact that later method is more accurate), but it has been written about -- try a set of values differing only in the last few of the 12 available significant digits. The fact is, that even computing the mean of two values as half of their sum, or (x+y)/2 can be less accurate than as 'x+(y-x)/2' (try with x=y=6.66666666667 for example) which may serve as a hint that all other statistics derived only from sums might not be as accurate as can otherwise be achieved point-by-point. > The disadvantage (is it so?) might be this situation, > in which I need to go on from these accumulated computed sums But you can still get all those sums (from array of actual data points), using built-in commands. Bob Wheeler provided other insights in the past, about how to better compute various statistics, as in the following case about weighted means and deviations: "Entering frequencied statistical data?" [November 1996] http://groups.google.com/group/comp.sys.hp48/browse_thread/thread/babb978de465a026/ [r->] [OFF]
From: lacro on 16 Jun 2008 23:06
Fine, tks John. I understood that point even from before. It will always be more accurate to calculate those figures WITH the Data Points, (and even more AFAIK even depending on the algorithms, I could loose accuracy). In any case, the situation is not that I prefer to use the sums directly, BUT that I don't have that chance in this particular case. I think I didn't express myself correctly before. The situation provided me only with those 6 sums (X, Y, XY, X2, Y2 and N), BUT couldn't LET me GET those actual data points (that would have allowed me to compute those very sums obviously better). I mentioned a "reference" because I assumed you'd probably refer my inquiry to a previous thread or some HP document where it is stated what you explained. However I will check Bob Wheelers thread as suggested. Regards, Carlos On Jun 16, 8:06 pm, "John H Meyers" <jhmey...(a)nomail.invalid> wrote: > On Sat, 14 Jun 2008 16:08:56 -0500, lacro wrote: > > I couldn't find that very interesting distintion reference you mentioned.. > > I didn't provide a reference (for the fact that later method is more accurate), > but it has been written about -- try a set of values differing only in the last few > of the 12 available significant digits. > > The fact is, that even computing the mean of two values as half of their sum, > or (x+y)/2 can be less accurate than as 'x+(y-x)/2' > (try with x=y=6.66666666667 for example) > which may serve as a hint that all other statistics derived only from sums > might not be as accurate as can otherwise be achieved point-by-point. > > > The disadvantage (is it so?) might be this situation, > > in which I need to go on from these accumulated computed sums > > But you can still get all those sums (from array of actual data points), > using built-in commands. > > Bob Wheeler provided other insights in the past, > about how to better compute various statistics, > as in the following case about weighted means and deviations: > > "Entering frequencied statistical data?" [November 1996]http://groups.google.com/group/comp.sys.hp48/browse_thread/thread/bab... > > [r->] [OFF] |