matplotlib-users

[Matplotlib-users] transformations

[N. V. <mit...@we...>]

Hello list,

I have just figured out how to use transformations for the 'Text' object 
from the class library to add a text to the graph (well actually this 
applies to any Artist object).

If the coordinates are given as absolute _screen_ coordinates (very 
unlikely...), we use:

  t = Text(x=x, y=y, text=text)
  axes.texts.append(t)

If however the coordinates are given as _data_ coordinates x,y then we use:

  t = Text(x=x, y=y, text=text)
  t.set_transform(axes.transData)
  axes.texts.append(t)

My question: Is there any transformation that would allow me to specify 
the coordinates as x,y tuple where (0,0) is bottom left and (1,1) is top 
right? In gnuplot this is known as the 'graph' coordinate system.

Best regards,

Niklas.

[Matplotlib-users] transformations

[Nicolas <nic...@ya...>]

Hello,

I tried to use some affine transformations, but it didn't match to get it works.
I use matplotlib 0.87.5 with python 2.4.3 and Numeric 24.2 on winXP.

I tried for example :
plot([1,0,1],[0,1,0], transform = matplotlib.transforms.Affine(0,1,1,0,0,0))

Then I got :
Traceback (most recent call last):
  File "<input>", line 1, in ?
  File "C:\Python24\lib\site-packages\matplotlib\pylab.py", line 2019, in plot
    ret =  gca().plot(*args, **kwargs)
  File "C:\Python24\lib\site-packages\matplotlib\axes.py", line 2124, in plot
    self.add_line(line)
  File "C:\Python24\lib\site-packages\matplotlib\axes.py", line 879, in add_line
    xys = self._get_verts_in_data_coords(
  File "C:\Python24\lib\site-packages\matplotlib\axes.py", line 934, in _get_verts_in_data_coords
    xys = trans.numerix_xy(asarray(xys))
ValueError: Domain error on Transformation::numerix_xy
>>> plot([1,0,1],[0,1,0], transform = matplotlib.transforms.Affine(1,0,0,1,0,0))
Traceback (most recent call last):
  File "<input>", line 1, in ?
  File "C:\Python24\lib\site-packages\matplotlib\pylab.py", line 2019, in plot
    ret =  gca().plot(*args, **kwargs)
  File "C:\Python24\lib\site-packages\matplotlib\axes.py", line 2124, in plot
    self.add_line(line)
  File "C:\Python24\lib\site-packages\matplotlib\axes.py", line 879, in add_line
    xys = self._get_verts_in_data_coords(
  File "C:\Python24\lib\site-packages\matplotlib\axes.py", line 934, in _get_verts_in_data_coords
    xys = trans.numerix_xy(asarray(xys))
ValueError: Domain error on Transformation::numerix_xy

Thanks a lot,

Nicolas
 
 		
---------------------------------
 Découvrez une nouvelle façon d'obtenir des réponses à toutes vos questions ! Profitez des connaissances, des opinions et des expériences des internautes sur Yahoo! Questions/Réponses.

Re: [Matplotlib-users] transformations

[John H. <jd...@gm...>]

On 2/1/07, Nicolas <nic...@ya...> wrote:

> I tried to use some affine transformations, but it didn't match to get it
> works.
> I use matplotlib 0.87.5 with python 2.4.3 and Numeric 24.2 on winXP.
>
> I tried for example :
> plot([1,0,1],[0,1,0], transform = matplotlib.transforms.Affine(0,1,1,0,0,0))

You cannot pass scalars to the Affine constructor -- matplotlib uses
value reference semantics in the transform, as explained at
http://matplotlib.sf.net/matplotlib.transforms.html and so you need to
construct your transformation like so:

import matplotlib.transforms as t

def make_affine(a,b,c,d,tx,ty):
    return t.Affine(*[t.Value(val) for val in (a,b,c,d,tx,ty)])


a = make_affine(0,1,1,0,0,0)

>>> print a.xy_tup((1,1))
(1.0, 1.0)

>>> print a.xy_tup((-1,1))
(1.0, -1.0)


Note there was a bug in older versions of mpl in the affine class that
was fixed, specifically in the sign of b and c.  I am not sure which
version that was off the top of my head, so if you see any problems
consider upgrading.

Re: [Matplotlib-users] RE : Re: transformations

[John H. <jd...@gm...>]

On 2/6/07, Nicolas <nic...@ya...> wrote:
> Thanks a lot for your reply.

You're welcome.  Please keep correspondence on the list so others an
contribute and benefit.

> I read this story of lazy value in the doc, but I didn't understand the link
> with the the Affine function.

All of the matplotlib transformations are by reference.  We want to be
able to do
build the transform like so:

  a = someval
  t  = Affine(a,b,c,d,e,f)

and then later in the code do

  a.set(23)

and have the transformation automatically updated.  You can't do this
if a is a scalar value (eg a float) so we have a custom class Value
which stores a pointer to a float.  When you do

  a = Value(23)
  t  = Affine(a,b,c,d,e,f)

and later do a.set(23) then t will be updated.  We also support
arithmetic on Values, so you can do

  z = Value(3)
  x = Value(2)
  t  = Affine(z*x,b,c,d,e,f)

and later if you do

  x.set(4)

then the transform will be updated to have the value a=12

>From a user perspective, it means you need do do

  t = Affine(Value(a), Value(b), .....)

if a and b are floats.  This is what the make_affine function that I
posted does.

> Additionnaly, I noticed that antialiasing doesn't seem to work anymore in
> function pie with recent (ie. 0.87.5) version of matplotlib on WxAgg (?)

Odd, I see that too.  Haven't been able to pinpoint the problem yet...

JDH

Re: [Matplotlib-users] RE : Re: RE : Re: transformations

[John H. <jd...@gm...>]

On 6/6/07, Nicolas <nic...@ya...> wrote:

> Unfortunately, I didn't use instantiated Polygon items myself, so for now I
> can't figure which part of my code produces indirectly such empty
> Polygons...
>
> May it be however possible to modify a little matplotlib code, so as to
> assure the compatibility with the previous existing versions of matplotlib ?
> Something like :

I don't have a problem with this, but it just makes it easier for
latent bugs in your code to remain hidden.  Should we raise in the
Polygon constructor is you pass in an empty list of vertices?  Of
course, you can always set xy directly, so this won't catch all the
errors unless we use properties or traits.

JDH

Re: [Matplotlib-users] RE : Re: RE : Re: transformations

[Nicolas <nic...@ya...>]

I can't figure why Polygons are used even when I simply draw lines.
But I know at least why I use empty list of vertices. ;)
In fact, I use matplotlib for a dynamic geometry software (something like
GeoGebra), and systematic use of high-level functions of matplotlib would be
to slow for that purpose.
And yet, I didn't want to use too low-level functions, like the renderer's
ones, for I didn't want to depend too much of a particular renderer.

So, I created once a set of plots and fills for every object, and then I
modified its attributes . I used to set its vertices to an empty list when I
didn't want it to appear on the figure.
I suppose now I'd better use instead something like.
    my_plot._visible = False

Nevertheless, I think empty vertices should be supported here (or not
supported at all for all renderers - for in 0.90.1 it still works with tkagg
for example)

Nicolas



On 6/6/07, John Hunter <jd...@gm...> wrote:
>
> On 6/6/07, Nicolas <nic...@ya...> wrote:
>
> > Unfortunately, I didn't use instantiated Polygon items myself, so for
> now I
> > can't figure which part of my code produces indirectly such empty
> > Polygons...
> >
> > May it be however possible to modify a little matplotlib code, so as to
> > assure the compatibility with the previous existing versions of
> matplotlib ?
> > Something like :
>
> I don't have a problem with this, but it just makes it easier for
> latent bugs in your code to remain hidden.  Should we raise in the
> Polygon constructor is you pass in an empty list of vertices?  Of
> course, you can always set xy directly, so this won't catch all the
> errors unless we use properties or traits.
>
> JDH
>

Re: [Matplotlib-users] transformations

[John H. <jdh...@ac...>]

>>>>> "N" == N Volbers <mit...@we...> writes:

    N> Hello list, I have just figured out how to use transformations
    N> for the 'Text' object from the class library to add a text to
    N> the graph (well actually this applies to any Artist object).

    N> If the coordinates are given as absolute _screen_ coordinates
    N> (very unlikely...), we use:

    N>   t = Text(x=x, y=y, text=text) 
    N>   axes.texts.append(t)

    N> If however the coordinates are given as _data_ coordinates x,y
    N> then we use:

    N>   t = Text(x=x, y=y, text=text) 
    N>   t.set_transform(axes.transData)
    N>   ax.texts.append(t)

You can also pass the transform as a kwarg

  t = Text(x=x, y=y, text=text, transform=axes.transData) 
  ax.texts.append(t)

    N> My question: Is there any transformation that would allow me to
    N> specify the coordinates as x,y tuple where (0,0) is bottom left
    N> and (1,1) is top right? In gnuplot this is known as the 'graph'
    N> coordinate system.

There are four built-in transforms that you should be aware of; below
ax is an Axes instance and fig is a Figure instance

  matplotlib.transforms.identity_transform()  # display coords
  ax.transData     # data coords
  ax.transAxes     # 0,0 is bottom,left of axes and 1,1 is top,right
  fig.transFigure  # 0,0 is bottom,left of figure and 1,1 is top,right

There is nothing wrong with instantiating Text instances yourself and
adding them to the figure or axes manually as you did, but both the
Axes and Figure have helper methods "text" which define a default
transformation that can be overridden

So

  t = Text(x=x, y=y, text=text) 
  t.set_transform(axes.transAxes)
  ax.texts.append(t)

is equivalent to 

  ax.text(x, y, text, transform=ax.transAxes) 

The default transform for ax.text is ax.transData and the default
transform for fig.text is fig.transFigure.

Of course, you can define more general transformations, eg
matplotlib.transforms.Affine, but the four listed above arise in a lot
of applications.

If you have time and inclination to write a transformations tutorial
for the wiki, that would be great.

Thanks,
JDH

Re: [Matplotlib-users] transformations

[N. V. <mit...@we...>]

Hello John,


>
>There is nothing wrong with instantiating Text instances yourself and
>adding them to the figure or axes manually as you did, but both the
>Axes and Figure have helper methods "text" which define a default
>transformation that can be overridden
>
>So
>
>  t = Text(x=x, y=y, text=text) 
>  t.set_transform(axes.transAxes)
>  ax.texts.append(t)
>
>is equivalent to 
>
>  ax.text(x, y, text, transform=ax.transAxes) 
>
>  
>
ax.text was my starting point to understand the usage of Text(). For my 
application I need a very detailed control of the objects I create, this 
is why I prefer the manual approach.

>
>If you have time and inclination to write a transformations tutorial
>for the wiki, that would be great.
>
>  
>
Not tonight, but I will!


Thanks so much for your answer,

Niklas.