Hello Gaël,
First, thanks for your answer and sorry for the delay with mine. I'm
new to Python in general, and had a lot to figure out.
I used the code you linked to, with some modifications, particularly
because I'm dealing with directed graphs (asymmetric adjacency
matrix).
To get arrows, I made the tubes transparent and put a quiver3d on
top (i.e. 'in') them:
p, li { white-space: pre-wrap; }
def
plotarrows(offset):
posx = x[start_idx]
posy = y[start_idx]
posz = z[start_idx]
dirx = x[stop_idx] -
posx
diry = y[stop_idx] -
posy
dirz = z[stop_idx] -
posz
# push the start of
the arrows away from the node sphere centres
# (offset = 0.5 means
they _start_ mid-way the tube)
posx = posx +
offset*dirx
posy = posy +
offset*diry
posz = posz +
offset*dirz
arrows =
mlab.quiver3d(posx, posy, posz, dirx, diry, dirz,
# sqrt() boosts
smaller weights visually
scalars =
np.sqrt(edge_scalar),
scale_mode =
'scalar',
scale_factor =
0.0075,
mode = 'arrow',
resolution = 6,
color = (1, 0, 0))
return arrows
That results in something like [1].
I tried to go fancy and animate the arrows, by creating three
quiver3ds, at different offsets, and switching them on/off (like a
traffic light):
p, li { white-space: pre-wrap; }
<at> mlab.animate(delay=10)
def anim():
ar = []
ar.append(plotarrows(0.1)); ar[0].stop()
ar.append(plotarrows(0.4)); ar[1].stop()
ar.append(plotarrows(0.7)); ar[2].stop()
while 1:
for i in [0, 1, 2]:
ar[i-1].stop()
ar[i].start()
yield
This works, but too slowly for practical use with large graphs (even
when the delay is set at 10ms - the true delay is much larger on my
system). I tried some alternatives: (1) using _hideshow() instead of
stop() and start(); (2) creating only one quiver3d, and updating its
x, y and z parameters while cycling through the three offsets, using
mlab_source as explained in [2]. I didn't get a noticeable speed
increase. Any alternatives for this? I haven't quite figured out the
details of the pipeline system yet...
Best regards,
Andreas
[1]
http://imageshack.us/photo/my-images/683/testsnapshot.png/
[2]
http://github.enthought.com/mayavi/mayavi/mlab_animating.html?highlight=animating%20data
On 06/12/11 07:08, Gael Varoquaux wrote:
On Sun, Dec 04, 2011 at 03:47:33PM +0100, andreasl <at> club.lemonde.fr wrote:
[1] I would like to make the diameter of each graph edge (tube) depend on an associated weight value. My impression is that tube_radius can only be a single value, not a vector, so I imagine a loop is required. Is that correct?
We can do slightly better, but it is non trivial. I posted some code to
do this recently on the enthought-dev mailing list (where many of the
Mayavi related discussions happen):
https://mail.enthought.com/pipermail/enthought-dev/2011-November/030194.html
You will need to study it and do some reading of the Mayavi documentation
on the pipeline and the difference modules to understand it.
RSS Feed