clusterviz-devel — This list is intended for developmental related issues of the ClusterViz package

[Clusterviz-devel] Re: Karsten's panning proposal

[Janne G. <gr...@mo...>]

> Hey Karsten and the rest of the world,
>
> I have noticed a few issues with your proposal.
>
> * Generally we want a simpler panning mechanism than what you thought
> of. The idea was just to pan in the plane perpendicular to the
> viewing direction. When I looked at your coded it occurred to me that
> this really should not be too difficult to implement. Basically it
> should suffice to (push another matrix, (maybe load identity - but
> probably not,) and then) do the translation in x-y-direction as
> dictated by the user's mouse movements (and pop matrix afterwards).
> One point that I did not think through entirely is when to do this
> manipulation - at the beginning or at the end of the matrix stack?

It is not that simple. I tried such approaches and the genral problem is 
that they will not work together with rotations. You have to do a 
translation in the viewing plane and not in the x-y-plane.
We want a x-y movement of the point of view. Since the camera has always 
coordinates 0,0,0 we must convert this camera translation into 
translation of the world.

> One complication with the Patches-service at SF is at the moment that
> it is not clear to exactly which revision a patch might apply as we
> don't make use of the SF-CVS in favor of our own private SVN-server
> to which you unfortunately don't have access (and it is not clear
> whether I can get you access there - though I will check). I need to
> change this situation. I will try and replicate the history of the
> SVN-server we are currently using into my own private one where it
> would be easy to grant you or other developers the necessary access
> rights until we make the transition to the hopefully soon fully
> functional SVN-service at SF.

Different Solution. We can add at the original svn server accounts just 
for the repository. I will add an acount for Karsten this afternoon and 
mail him instructions.

ciao Janne

[Clusterviz-devel] Karsten's panning proposal

[Jonas M. <na...@us...>]

Hey Karsten and the rest of the world,

I have noticed a few issues with your proposal.

* you make use of a self-written function "inner_prod". There is of course 
nothing wrong with that. I only want to draw your attention to the presence 
of the function Numeric.innerproduct (for reference see a couple of example 
usages in visualkmeans.py). The Numeric package is a library written in C and 
provides Python bindings. This implementation is fairly efficient, thus my 
recommendation. But: development for Numeric has ceased and it's use is 
deprecated today in favor of its successor NumPy. We might one day make the 
switch to using NumPy but this is not of utmost importance as of today.

* Generally we want a simpler panning mechanism than what you thought of. The 
idea was just to pan in the plane perpendicular to the viewing direction. 
When I looked at your coded it occurred to me that this really should not be 
too difficult to implement. Basically it should suffice to (push another 
matrix, (maybe load identity - but probably not,) and then) do the 
translation in x-y-direction as dictated by the user's mouse movements (and 
pop matrix afterwards).
One point that I did not think through entirely is when to do this 
manipulation - at the beginning or at the end of the matrix stack? Could 
maybe work best by accumulating it into this .pan matrix that you are 
calculating in your code proposal.

* Another thing would be the following. If you have patches please use the 
sourceforge service to file them (for there is also a limit of 40KB on the 
size of mails on this list). If you go to the project sourceforge pages you 
will find a link labeled "Patches" in the menu where you can submit patches 
to the code. Here it would be sensible to commit real patches (like diffs) as 
opposed to replicating the whole non-changed code so that it is easier to 
spot the differences.
As a consequence I can not try out your code right now, as I am missing part 
of the clusterdisplay.py file. Please either resend to me privately or use 
the patch-submission service (you might think about patching the new revision 
that I sent you).

One complication with the Patches-service at SF is at the moment that it is 
not clear to exactly which revision a patch might apply as we don't make use 
of the SF-CVS in favor of our own private SVN-server to which you 
unfortunately don't have access (and it is not clear whether I can get you 
access there - though I will check). I need to change this situation. I will 
try and replicate the history of the SVN-server we are currently using into 
my own private one where it would be easy to grant you or other developers 
the necessary access rights until we make the transition to the hopefully 
soon fully functional SVN-service at SF.


Kind regards, Jonas.

[Clusterviz-devel] Prefered language English

[Jonas M. <jo...@ma...>]

Please use English on this mailling list. Even though there are only German 
speaking developers at the moment future developers might not be fluent in 
German. Thus the prefered language of conversation on this list is English.

Kind regards, Jonas Maaskola.

[Clusterviz-devel] trackball.py

[<K.W...@gm...>]

Hallo,
anbei ein Vorschlag für den Filmschwenk..

Sobald die mittlere Maustaste gedrückt wird, wird mit _gltbPointToVector den
Mauskoordinaten
ein Punkt auf dem Trackball zugeordnet. Dann wird die Ebene berechnet,
die den Trackball in diesem Punkt berührt. Jede folgende Mausbewegung bei
gedrückter
Maustaste wird als Bewegung in dieser Ebene interpretiert.

Wirklich gut finde ich die Funktion noch nicht, aber bevor ich mich noch mal
daran setze,
würde ich gern Eure Meinungen hören. Geht das so in die Richtung, die ihr
Euch vorgestellt
habt?

Ich habe nicht nur die trackball.py, sondern auch clusterdisplay.py
geändert, wegen des ersten
Mausklicks. Dabei habe ich die Version von Sourceforge genommen, nicht die
"Projektionsversion".

Viele Grüße,
Karsten.

-- 
Telefonieren Sie schon oder sparen Sie noch?
NEU: GMX Phone_Flat http://www.gmx.net/de/go/telefonie

[Clusterviz-devel] trackball

[<K.W...@gm...>]

Hallo!
Inzwischen habe ich mir die trackball.py nochmal angesehen. Hier ein paar
Gedanken:

- Der (imaginäre) Trackball hat den Mittelpunkt (0,0,0) und Radius 1.

- Die Funktion _gltbPointToVector ordnet einer Bildschirmkoordinate (x,y) 
  einen Punkt auf der Oberfläche des Trackballs zu. Schöne Funktion :-)
  Das d<1.0 scheint mir recht willkürlich, mit d<1.4 funktioniert's auch.

- recalculateTransformationMatrix berechnet die neue Transformationsmatrix 
  T_neu wie folgt:
    T_neu = T_alt o Rotation(angle, axis). ("o" für Komposition)
  Ich hätte eigentlich erwartet, dass (in gltbMotion) zumindest beim 
  ersten Aufruf 
    transform*lastposition=current_position 
  gilt, aber es gibt da Abweichungen schon an der 3. Stelle hinterm 
  Komma, habe ich da was falsch verstanden?
  
- Das Problem beim Panning scheint mir zu sein, die Bewegung der Maus 
  als Bewegung im Raum zu interpretieren. Die Funktion _gltbPointToVector 
  ist wohl nicht geeignet, weil sie immer einen Norm-1 Vektor zurückgibt. 
  Was meint ihr dazu? Auch ist mir ist unklar, warum die dritte Zeile der 
  Transformationsmatrix die Blickrichtung sein soll...

Dann habe ich noch herausgefunden, dass es eine Funktion gluLookAt gibt, 
die ganz nützlich sein könnte und dass mit PyOpenGL ein Tkinter-Widget Togl
kommt, welches ähnliche Funktionen wie die trackball Klasse hat. 

Meine sourceforge-Kennung ist übrigens KarstenWeinert und wenn ich 
  svn co https://cvs.molgen.mpg.de/InfoTheory
eingebe, werde ich nach einem Benutzernamen gefragt, wobei "anonymous" nicht
akzeptiert wird. Die Projektionsversion konnte ich noch nicht probieren,
weil offenbar ein Modul auxiliary.py fehlt.

Eigentlich wollte ich mich ja noch letztes Jahr melden, aber irgendwie war
die Zeit zwischen den Feiertagen zu kurz. Naja um ehrlich zu sein, war ich
fauler als ich dachte... 

Jedenfalls wünsche ich Euch für 2006 alles Gute!

-- Karsten.

-- 
DSL-Aktion wegen großer Nachfrage bis 28.2.2006 verlängert:
GMX DSL-Flatrate 1 Jahr kostenlos* http://www.gmx.net/de/go/dsl