HP Forums

I'm using the App Statistics 2Var to do a quadratic fitting to the following data:

C1: 1, 1.05, 1.1, 1.14, 1.16
C2: 68, 67.2, 62.8, 59.1, 54.7

Once the fitting is done, using PredX and PredY in HOME I get:

PredY(1.1) -> 63.3070360465 OK
PredX(63.3070360465) -> 0.9 ??? I expected 1.1

After some investigation I realized that the fitting parabola is symmetrical with respect a vertical axis at x=1, so PredY(0.9)=PredY(1.1)=63.3070360465, and mathematically both PredX(63.3070360465)=0.9 and PredX(63.3070360465)=1.1 are correct solutions.

Perhaps in cases like this, when PredX has more than one solution, the calculator should favor a value that is inside the data range (1.1 in this case), instead of returning the first solution found (0.9 in this case).

In CAS you could do:

fsolve((eval(S1(5))) = 63.3070360465,X)

to get:

[1.1,0.915714357502]

-road

So extrapolation is not allowed? Isn't that the point of a "Pred"?

When implementing this we had some long discussions about what to do here and couldn't come up with a good solution. There really isn't a "nice" way to do this since you could just as easily want the "other" answer. It gets even more tricky when you support N degree polynomials or trig fits.

What we basically decided was that just "looking at the plot" was a perfectly acceptable solution.

Thanks for your answers.

So what Tim is essentially saying is that, as usual, you have to be smarter than the machine and not trust it blindly! Then using the method described by roadrunner one can get the right solution.

I'm not certain, but I think most calcs only offer the
"predx" type of capability and others don't do the "predy" at all.

(04-17-2017 05:10 PM)Tim Wessman Wrote: [ -> ]I'm not certain, but I think most calcs only offer the
"predx" type of capability and others don't do the "predy" at all.

Isn't it usually the opposite? PredY simply means you plug in the new X and see what the model returns.

Correct. I mis-typed there. Thanks!