Suggestion

29 views
Skip to first unread message

swharden

unread,
Jun 10, 2009, 2:40:45 PM6/10/09
to Zunzun.com discussion group
I hope this is the correct place to post suggestions. I've got to
say, your service is AWESOME! Python, numpy, and matplotlib work
wonderfully together, and incorporating their curve fitting
functionality into a web interface was an awesome idea. Of course I
could do all of this by hand in the code, but it's so much faster and
easier to use your service! Hats off to you!!!

I would recommend, however that you provide representative images (or
even links to descriptions) next to the most basic categories of
equations. Visually distinguishing your desired curve formula from
representative classes would greatly simplify the of selecting which
formula you want. i.e., normal/inverse exponential/logarithmic with/
without linear decay curves, etc.

zunzun.com

unread,
Jun 10, 2009, 3:11:09 PM6/10/09
to Zunzun.com discussion group
Excellent suggestion. I will add this to my "TO_DO" list.

James

Plantiful

unread,
Jul 6, 2009, 4:10:47 PM7/6/09
to Zunzun.com discussion group
This webservice is going to be a key site for anyone that is curve-
fitting-- a colleague showed this to me, and I am very impressed with
the information that can be calculated here! Great job!

We have a data set that actually fits the Quadratic Formula, y = (-b +
pow(pow(b,2)-4*a*(c-x)),0.5)/2/a, extremely well. We found this
initially by inverting the axes and fitting to a second-order
polynomial, but this does not fit the lower-values very well (the
intercept error is large compared to the actual value). In fitting
the (x,y) data to the quadratic formula, using the LMFIT program
(Levenberg-Marquardt), we get near-perfect results. I have included
the data below:

X Y
1000 507912
500 335301
200 184001
100 109480
50 62661
20 28863
10 14354
5 6551

Can the Quadratic formula be included as a possible equation, or did I
miss it?

zunzun.com

unread,
Jul 7, 2009, 10:25:48 AM7/7/09
to Zunzun.com discussion group
I can easily add a new equation. Somehow the parentheses don't look
right in the expression you posted, can you please check it?

James

On Jul 6, 3:10 pm, Plantiful <marcplant...@gmail.com> wrote:
> We have a data set that actually fits the Quadratic Formula, y = (-b +
> pow(pow(b,2)-4*a*(c-x)),0.5)/2/a, extremely well.
>

Marc Plante

unread,
Jul 7, 2009, 11:07:01 AM7/7/09
to zunzun_...@googlegroups.com
Hi James,

You were correct, and the equation is below:

y = (-b + pow(pow(b,2) - 4 * a * (c - x), 0.5)) / 2 / a

Thank you,
Marc

zunzun.com

unread,
Jul 7, 2009, 11:55:00 AM7/7/09
to Zunzun.com discussion group
This is now in the 2D Polynomials, you can't miss it. Please test it.

James

Marc Plante

unread,
Jul 7, 2009, 12:30:41 PM7/7/09
to zunzun_...@googlegroups.com
No problem finding it!  The equation works as intended.

Thank you for your work on this webpage and site!

Marc

zunzun.com

unread,
Jul 7, 2009, 1:46:20 PM7/7/09
to Zunzun.com discussion group
Excellent. You now have an entry on the home page "Hall Of Fame", and
my thanks as well.

James

Marc Plante

unread,
Jul 7, 2009, 2:00:03 PM7/7/09
to zunzun_...@googlegroups.com
I am simply honored!

Will you be homogenizing the equation-format to be consistent with the rest of your equations?
This fit is not really custom (this will make people nervous, LOL). 

If my colleague in France finds this site useful, we may include it in a journal publication, but this will take some time.

Cheers,
Marc

zunzun.com

unread,
Jul 7, 2009, 2:11:01 PM7/7/09
to Zunzun.com discussion group
On Jul 7, 1:00 pm, Marc Plante <marcplant...@gmail.com> wrote:
> Will you be homogenizing the equation-format to be consistent with the rest
> of your equations?

No. It stays as you request it - you know what you need better than I
do.


> This fit is not really custom (this will make people nervous, LOL).

Nah. "Custom" is just a name, and I think it shows people I'm clearly
willing to add request equations,


> If my colleague in France finds this site useful, we may include it in a
> journal publication, but this will take some time.

Cool beans!


James

Marc Plante

unread,
Jul 7, 2009, 2:18:08 PM7/7/09
to zunzun_...@googlegroups.com

OK.  Please substitute this equation then, as few people will understand the "pow( )" function.  I could not do exponents by gmail, so I typed it in Word and copied it.  (the first '2' and the '0.5' are exponents).

y = (-b + (b2 – 4 a (c – x))0.5) / 2 / a

Thanks,
Marc

zunzun.com

unread,
Jul 7, 2009, 2:37:15 PM7/7/09
to Zunzun.com discussion group
Done.

James

Marc Plante

unread,
Jul 7, 2009, 2:39:55 PM7/7/09
to zunzun_...@googlegroups.com
Looks great, James!

zunzun.com

unread,
Jul 7, 2009, 2:44:38 PM7/7/09
to Zunzun.com discussion group
I think you have now seen in real time one of the advantages to having
this software in the form of a web site.

James

On Jul 7, 1:39 pm, Marc Plante <marcplant...@gmail.com> wrote:
> Looks great, James!

Marc Plante

unread,
Jul 7, 2009, 3:07:54 PM7/7/09
to zunzun_...@googlegroups.com
Indeed-  it is also much easier than building custom code or adapting generic code to fit an application.  I sent my colleague, who is far more knowledgeable in statistics than I, three runs to see how he likes it to see if we can get some publicity for you.

Marc

Robin

unread,
Jul 30, 2009, 8:31:45 AM7/30/09
to Zunzun.com discussion group
Excellent website. Very impressed.
Was wondering if you have any plans for extending to beyond 3D
response surfaces. I can think of a few characterization problems I
have coming up with 3 or more independent variables.

Robin

zunzun.com

unread,
Jul 30, 2009, 9:36:41 AM7/30/09
to Zunzun.com discussion group
There is no theoretical or programmatic barrier to doing so, but I
have no current plans for this - the problem is in the presentation of
the graphs and visualizations over the web. 2D is OK, 3D is doable -
but 4D, 5D or 6D? I don't think higher data dimensions are practical
using a web interface given current Internet technology.

James
Reply all
Reply to author
Forward
0 new messages