The LIST

[chilly_charly]

Ben,

Nice to see that you jumped on the subject.

Unfortunately, during a few days, my only possible way to use 3D graphs, will be to run LV 7.0 with a PC emulator on a Macintosh... Could you post a version of your patch-work as 7.0 ?

CC

[Ben]

Code as LV 7.0 with smaller 3-d graph.

Also included is jpg of big 3-d graph that illustrates four atractors (Inactive, All buisness, Consumers, Geeks asper my wife).

The attached code and plot prompts me to question what I'd see if I took all of the old lists, and plotted all of us progressing through your PCA sapce over time.

The attractors also imply I colud calculate the center of mass for each region to get a theoretical "typical geek" and then use statistics to bin each user (based on stdev of displacement in PCA space) with transparent spheres to deliniate boundries.

What do you think?

Ben

Message Edited by Ben on 06-05-2005 09:39 AM

[Ben]

I wen back and looked and it appears you have normilized. The following was written before I looked again.


No, I did not read your extra link yet.

Is there a way we can scale the various axis in a meaningful fashion?

I pose the Q because if I were to do my center of mass calculations based on the current values it seems dispalcements in one dimension could dwarf the others. Is this already taken into account?

Does doing the center of mass in a logarithmic space get complicated?

Let me stop and comend you on this application CC!

I am impressed how you took this mind boggling pile of numbers and managed to squeeze out what appears to be something meaningful.

I am going to show my math weakness here so brace yourself.

It looks like your PCA stuff is a type of convolution. If we found the C.O.M. for each attractor and then deconvolved them we would then have the measurements of the typical geek in terms of the measurements that show up in your list.

Lots of Q's. So little time to play.

Ben

Message Edited by Ben on 06-05-2005 09:59 AM

[Ben]

Do you tink we should take this topic to the LV discusion forum?

We may get some more input from the "buisness only" types.

Ben

[chilly_charly]

Ben,

STOP 😉

Ououf ! Just give me some time to study your artwork and prepare a answer ! I'have just downloaded your artwork.

Calculating the center of mass of the aggregates is of course a nice idea, and the logical next step.
However, Principal Component Analysis is not the best tool to do that. There are other method that could give a much better discrimination, for instance Factorial Discriminant Analysis (FDA). There are even automatic classification methods that could be used.

Let me first study your findings, and I'll come back later.

Thanks for the excitation and the fun. Clearly this seems to be something new for you, but I feel that you have already seen a number of possible applications for PCA. And indeed there are...

CC

Message Edited by chilly charly on 06-05-2005 07:38 PM

[chilly_charly]

Ben

I have opened the last version. As usual, some cosmetic problems with the 3D graph (obliged to re-configure the display).

The problem is that I don't see where you made the modifications. And specifically where you changed the scales to log. Could you give me some light ?

CC

[Ben]

I just went to the

CWProperties >>> Axes >>> Range

and selected Log.

Here it is again complete with explicit setting of axes for log as 7.0.

Sorry about the bandwidth.

Yes sir, this is new stuff to me. My math ended about 2-3 lectures after fractals were mentioned.

Ben

[chilly_charly]

---

@Ben wrote:
I just went to the

CWProperties >>> Axes >>> Range

and selected Log.
Ben

---

Yes, I see. But you were not allowed to do that : you can't calculate the log of negative numbers !
You should get 5 negative stars for that basic mistake ! 😄 😄 😄
Seems that the 3D graph is fault tolerant, and use the smallest possible value when asked to plot an "imaginary" log, without any warning. Should we report this as a bug ?

So, you simply separated the negative and positive scores !..

This should not change a lot the way you perceived the situation, but the nice aggregates have just vanished... 😞

Some improvements can be obtained by selecting a different set of variables, in the get data&class sub vi.
I'll try to spend some time on that...

CC

[chilly_charly]

I'm terribly sorry 😞

No List today, my PC went away...

I had not enough time to adapt the application to my Mac. You'll have to wait until next week...

[chilly_charly]

Back to "work".

With a 2-week break, the List evolution is more perceptible.

tst progress has been impressive, with a 4 places jump in the list. He/she/it will probably reach the 5th position during next week. No further fight to be foreseen in a next future, since altenbach is climbing also at an impressive rate.
Assuming the climbing rates remain constant, ben can relax for about 10 months before being seriously challenged by altenbach.
Ray seems to recover and passed the 1K "true replies" mark.

Down in the list, NI members are progressing more rapidly than non-NI members. Some blue blocks are remarquable. See for instance the range 200-250, 70% blue.

But the most active members are still non-NI (hurrah !..). Just sort the List according to last month activity, and spot the future top-enthusiats !..

Welcome to 8 new members : Japan_, rberger, AL, Alex D, Odd_Modem, bmihura, shahina, Nick C.

The List lodges now 479 members, with 196 NI enthusiasts.