matplotlib-users Mailing List for matplotlib

On Sun, 25 Nov 2007, Angus McMorland wrote:

> I'm not completely sure, but I suspect that this is an implementation bug,
> rather than a version bug, particularly because the line in question isn't
> involving matplotlib at all. If you post the relevant code
> (normal-curve.py, by the looks of things), it might be easy to spot the
> problem.

Angus,

   I've seen the same error trying to plot other curves in the past couple of
days, but not those using the Boltzmann distribution. Here's the file:

import matplotlib.numerix as nx
import pylab as p
from math import *

center = 50.0
fwhm = 50.0

def fwhm2k(fwhm):
   '''converts fwhm value to k (see above)'''
   return fwhm/(2 * nx.sqrt(nx.log(2)))

def gauss1d(r, fwhm, center):
   '''returns the 1d gaussian given by fwhm (full-width at half-max),
   and c (centre) at positions given by r
   '''
   return exp(-(r-center)**2 / fwhm2k(fwhm)**2)

x = nx.arange(0, 100, 0.1)
G = gauss1d(x, fwhm, center)
p.plot(x, G, color='red', lw=2)
p.axis([0, 100, 0.0, 1.0])
p.xlabel('Universe of Discourse')
p.ylabel('Membership Grade')
p.show()

> I've found it easiest to solve these sorts of bugs by running the code in
> an ipython shell, with automatic pdb calling. That way you can inspect the
> values of the parameters in question - one of which is, I think, the
> problem here.

   I've not run ipython with pdb; I'll look at the docs to learn how. I do
use winpdb on the application.

Thanks,

Rich

-- 
Richard B. Shepard, Ph.D.               |  Integrity            Credibility
Applied Ecosystem Services, Inc.        |            Innovation
<http://www.appl-ecosys.com>     Voice: 503-667-4517      Fax: 503-667-8863

On 25/11/2007, Rich Shepard <rsh...@ap...> wrote:
> On Sat, 24 Nov 2007, Angus McMorland wrote:
>
> > Great. Hopefully this correction will make things even more clear.
>
>    While the functions and equations are now clear, I get an error that was
> present in matplotlib-0.87, but which should be fixed in -0.90.1:
>
> Traceback (most recent call last):
>    File "normal-curve.py", line 19, in ?
>      G = gauss1d(x, fwhm, center)
>    File "normal-curve.py", line 16, in gauss1d
>      return exp(-(r-center)**2 / fwhm2k(fwhm)**2)
> TypeError: only length-1 arrays can be converted to Python scalars
>
>    According to /usr/lib/python2.4/site-packages/numpy/version.py, the
> installed version is 1.0b5.
>
>    Do I have a version incompatibility here?

I'm not completely sure, but I suspect that this is an implementation
bug, rather than a version bug, particularly because the line in
question isn't involving matplotlib at all. If you post the relevant
code (normal-curve.py, by the looks of things), it might be easy to
spot the problem.

I've found it easiest to solve these sorts of bugs by running the code
in an ipython shell, with automatic pdb calling. That way you can
inspect the values of the parameters in question - one of which is, I
think, the problem here.

Angus.
-- 
AJC McMorland, PhD Student
Physiology, University of Auckland

Hi,

I was able to make Scatter3D erroneously shift its plot down by 0.2. I
have no idea why this happened but I can reproduce it.

Basically I have data ranging from 0.0 to -1.0 over an x,y grid but
Scatter3D displays the data shifted down by 0.2!!!

I have included some screen shots of this problem, a python program
which creates it, and the data that I used.

Any help that could be given would be much appreciated.

Karl Edler

On Sat, 24 Nov 2007, Angus McMorland wrote:

> Great. Hopefully this correction will make things even more clear.

   While the functions and equations are now clear, I get an error that was
present in matplotlib-0.87, but which should be fixed in -0.90.1:

Traceback (most recent call last):
   File "normal-curve.py", line 19, in ?
     G = gauss1d(x, fwhm, center)
   File "normal-curve.py", line 16, in gauss1d
     return exp(-(r-center)**2 / fwhm2k(fwhm)**2)
TypeError: only length-1 arrays can be converted to Python scalars

   According to /usr/lib/python2.4/site-packages/numpy/version.py, the
installed version is 1.0b5.

   Do I have a version incompatibility here?

Rich

-- 
Richard B. Shepard, Ph.D.               |  Integrity            Credibility
Applied Ecosystem Services, Inc.        |            Innovation
<http://www.appl-ecosys.com>     Voice: 503-667-4517      Fax: 503-667-8863

On Nov 23, 2007 2:00 PM, Mark Bakker <ma...@gm...> wrote:
> Arnar -
>
> I presume you are using Tk.
> This is a known bug under Tk, that is apparently difficult to solve.
> John Hunter looked into it a year back or so, but couldn't find help
> from Tk experts that may know the solution.
> Oddly enough, I have written my own GUI in Tk, where I modified the
> toolbar, and I don't have the problem.
>
> If this could be solved, I would be very happy,

I'm pretty sure this is bug w/ tkagg is fixed in svn, because we
reverted to the old file save dialog that was causing it.  I am
surprised to hear that it is on gtk or gtkagg, since I've never hear
any reports of this.  In any case 0.87 is over a year old now, so I
wonder if Arnar could upgrade.

JDH

On Sat, 24 Nov 2007, Angus McMorland wrote:

> Looking at my reply, I realised this was rubbish - sorry about that. The
> fwhm is the difference between the two values of x that give Y = 0.5.

   Now that makes much more sense. Having control over the x values for the
inflection point allows us to tune the curve shape to more precisely reflect
the underlying semantics.

>>    The additional explanation helps a lot.
>
> Great. Hopefully this correction will make things even more clear.

   I had scanned your previous response without looking deeply at it because
I was not going to work on this until today in any case.

   For anyone curious about the context for my questions, it is an
approximate reasoning (i.e., expert system) model using fuzzy sets and fuzzy
logic. You can read all about it in my book, "Quantifying Environmental
Impact Assessments Using Fuzzy Logic," published in 2005 by Springer-Verlag.

Rich

-- 
Richard B. Shepard, Ph.D.               |  Integrity            Credibility
Applied Ecosystem Services, Inc.        |            Innovation
<http://www.appl-ecosys.com>     Voice: 503-667-4517      Fax: 503-667-8863

Hi
a quick note abut a "super package":
to let people know that now OpenSuse 10.3 has a set of RPMs for all scipy,
numpy, mpl, ipython, and others some of these being on the "science" repository,
e.g.:

http://download.opensuse.org/repositories/science/openSUSE_10.3/x86_64/

for the 64b distrib (ipython is included in the DVD or ftp main site)

working flawlessly (so far) for me. 10.3 seems to have been quite a nice version
(I can't say that for the older versions, and I have been with Suse since < v7)

Eric

A question was posted a week or so ago on the scipy list about installing
Python/mpl/numpy/scipy/Ipython by David Arnold (both Windows and Mac). He
was interested in a simple install for students (and others that are less
computer-savy). We had a little off-list correspondence, but thought to
bring it back to the list, also posting on mpl (cause I read that list and
it seems pretty responsive) and the numpy Google discussion group (they
respond VERY quickly).

On the scipy list it was suggested to use Sage reply, but that sounds like a
little overkill to me.

I have used the enthought package on windows successfully in class in
the past, but I understand that they have now switched to eggs, which
is not convenient (yet).

I also don't like all the other things I have to set myself. A switch
on the IDLE GUI to get mpl to work interactively (which is now seems
impossible to do), and setting my PYTHONPATH with environment variables. It
is all easy to do, but not for students, especially students without much
knowledge of how computers work let alone access to set these variables and
options.

At one point I was reading about making a superpack (or a name like
that) installation for Python with our favorite packages as mentioned above.
I am not sure where that is going.

Are others dreaming about such a superpack? What needs to be done to get it
going?

Mark

Well, that is very odd. Do Tk and GTK have the same problem? Was this known?

Is there a more substantial problem here than a backend problem?
Mark

On Nov 24, 2007 11:31 AM, Arnar Flatberg <arn...@gm...> wrote:

> Actually, I am using the GTK (Agg) backend. But thanks for replying.
>
> Arnar
>
> On Nov 23, 2007 9:00 PM, Mark Bakker <ma...@gm...> wrote:
> > Arnar -
> >
> > I presume you are using Tk.
> > This is a known bug under Tk, that is apparently difficult to solve.
> > John Hunter looked into it a year back or so, but couldn't find help
> > from Tk experts that may know the solution.
> > Oddly enough, I have written my own GUI in Tk, where I modified the
> > toolbar, and I don't have the problem.
> >
> > If this could be solved, I would be very happy,
> >
> > Mark
> >
> >
> > > From: "Arnar Flatberg" <arn...@gm...>
> > > Subject: [Matplotlib-users] After saving in GUI,        how can I get
> the
> > >         oo-interface?
> > > To: mat...@li...
> > > Message-ID:
> > >         <5d3...@ma...>
> > > Content-Type: text/plain; charset=ISO-8859-1
> >
> > >
> > > Hi list
> > >
> > > It looks like I cant change any properties of a matplotlib figure/axis
> > > after I saved the figure
> > > using the GUI. At least I cant use gca(), and gcf(). It works fine if
> > > the save (e.g., figure.savefig('foo'))
> > > was performed in the oo interface
> > >
> > > Does anybody else have a problem with this (code below)?
> >
>

Actually, I am using the GTK (Agg) backend. But thanks for replying.

Arnar

On Nov 23, 2007 9:00 PM, Mark Bakker <ma...@gm...> wrote:
> Arnar -
>
> I presume you are using Tk.
> This is a known bug under Tk, that is apparently difficult to solve.
> John Hunter looked into it a year back or so, but couldn't find help
> from Tk experts that may know the solution.
> Oddly enough, I have written my own GUI in Tk, where I modified the
> toolbar, and I don't have the problem.
>
> If this could be solved, I would be very happy,
>
> Mark
>
>
> > From: "Arnar Flatberg" <arn...@gm...>
> > Subject: [Matplotlib-users] After saving in GUI,        how can I get the
> >         oo-interface?
> > To: mat...@li...
> > Message-ID:
> >         <5d3...@ma...>
> > Content-Type: text/plain; charset=ISO-8859-1
>
> >
> > Hi list
> >
> > It looks like I cant change any properties of a matplotlib figure/axis
> > after I saved the figure
> > using the GUI. At least I cant use gca(), and gcf(). It works fine if
> > the save (e.g., figure.savefig('foo'))
> > was performed in the oo interface
> >
> > Does anybody else have a problem with this (code below)?
>

On 24/11/2007, Rich Shepard <rsh...@ap...> wrote:
> On Sat, 24 Nov 2007, Angus McMorland wrote:
>
> > fwhm is the full-width at half the maximum height, i.e. it's the
> > difference between the two values of x when:
> >
> > |r - c| = 0.5

Looking at my reply, I realised this was rubbish - sorry about that.
The fwhm is the difference between the two values of x that give Y =
0.5.

>    The additional explanation helps a lot.

Great. Hopefully this correction will make things even more clear.

Angus.
-- 
AJC McMorland, PhD Student
Physiology, University of Auckland

On Sat, 24 Nov 2007, Angus McMorland wrote:

> fwhm is the full-width at half the maximum height, i.e. it's the
> difference between the two values of x when:
>
> |r - c| = 0.5

Angus,

   The additional explanation helps a lot.

> The fwhm is a shape parameter (like std dev) - it determines the width of
> the curve. The combination of width and the range of values you plot (r)
> determine how close the function gets to zero, and how much of it is
> plotted. As Jeff said, it'll never actually reach zero, so you have to
> decide how close is close enough.
>
> You don't need to call fwhm2k yourself;  it's called by the gauss1d
> function. I just do it that way because the equation uses k, but I'm
> always interested in fwhm.

   Ah, I missed seeing that.

> Perhaps the easiest thing is to shove it into some quick code and play
> around with the values so you see how it works:

   That's what I intend to do. I've been running ipython and also writing
small scripts to understand how the functions work ... and plot. Playing
with parameters clarifies everything.

Much appreciated,

Rich

-- 
Richard B. Shepard, Ph.D.               |  Integrity            Credibility
Applied Ecosystem Services, Inc.        |            Innovation
<http://www.appl-ecosys.com>     Voice: 503-667-4517      Fax: 503-667-8863

On 24/11/2007, Rich Shepard <rsh...@ap...> wrote:
> On Fri, 23 Nov 2007, Angus McMorland wrote:
>
> > For parsimony, I think you're probably best off just using the Gaussian
> > equation:
> >
> > def fwhm2k(fwhm):
> >    '''converts fwhm value to k (see above)'''
> >    return fwhm/(2 * n.sqrt( n.log( 2 ) ) )
> >
> > def gauss1d(r, fwhm, c):
> >    '''returns the 1d gaussian given by fwhm (full-width at half-max),
> >    and c (centre) at positions given by r
> >    '''
> >    return exp( -(r-c)**2 / fwhm2k( fwhm )**2 )
> >
> > (released to public domain)
>
> Angus,
>
>    I'm trying to find the context for the above so I know what to feed fwhm2k
> as the fwhm value.

fwhm is the full-width at half the maximum height, i.e. it's the
difference between the two values of x when:

|r - c| = 0.5

The fwhm is a shape parameter (like std dev) - it determines the width
of the curve. The combination of width and the range of values you
plot (r) determine how close the function gets to zero, and how much
of it is plotted. As Jeff said, it'll never actually reach zero, so
you have to decide how close is close enough.

You don't need to call fwhm2k yourself;  it's called by the gauss1d
function. I just do it that way because the equation uses k, but I'm
always interested in fwhm.

>    It's been decades since I needed to work with continuous distributions and
> my insights and skills have rusted.

Perhaps the easiest thing is to shove it into some quick code and play
around with the values so you see how it works:

import pylab as p, numpy as n
x = n.arange(100) - 50.
fwhm = 25
centre = 0
y = gauss1d(x, fwhm, centre)
p.plot(x,y)

If you have other questions, you'll need to be a bit more specific so
we can address them directly.

Angus.
-- 
AJC McMorland, PhD Student
Physiology, University of Auckland

On Thu, 22 Nov 2007, Rich Shepard wrote:

>> For parsimony, I think you're probably best off just using the
>> Gaussian equation:
>>
>> def fwhm2k(fwhm):
>>    '''converts fwhm value to k (see above)'''
>>    return fwhm/(2 * n.sqrt( n.log( 2 ) ) )
>>
>> def gauss1d(r, fwhm, c):
>>    '''returns the 1d gaussian given by fwhm (full-width at half-max),
>>    and c (centre) at positions given by r
>>    '''
>>    return exp( -(r-c)**2 / fwhm2k( fwhm )**2 )
>
>   Thank you, Angus. I'll look at the Gaussian explanation to understand the
> input values.

   It's been decades since I've needed to write any statistical or
distributional function code; I used SysStat in DOS for quite some time and
R for the past decade with linux. There's a great difference between using
working code and writing one's one code. :-)

   Looking more carefully at the SciPy stats module, I see the normal
distribution is included as a class. However, I don't understand all the
functions in that class -- other than normal.pdf(x) -- and do not see where
loc and scale (representing mean and std. deviation) are used. Toward
understanding how to use this code to draw the curves I need, I extracted
the one method into a stand-alone file and tried to plot a Normal/Bell curve
from 0 to 100:

import matplotlib.numerix as nx
import pylab as p
from math import *

def normal(x):
   return exp(-x**2.0/2.0)/sqrt(2.0*pi)

x = nx.arange(0, 100, 0.1)
N = normal(x)
p.plot(x, N, color='red', lw=2)
p.axis([0, 100, 0.0, 1.0])
p.xlabel('Universe of Discourse')
p.ylabel('Membership Grade')
p.show()

   The error returned is:

Traceback (most recent call last):
   File "normal-curve.py", line 17, in ?
     N = normal(x)
   File "normal-curve.py", line 14, in normal
     return exp(-x**2.0/2.0)/sqrt(2.0*pi)
TypeError: only length-1 arrays can be converted to Python scalars

   A clue stick to the meaning of this error message, and how to fix the
problem is needed.

Rich

-- 
Richard B. Shepard, Ph.D.               |  Integrity            Credibility
Applied Ecosystem Services, Inc.        |            Innovation
<http://www.appl-ecosys.com>     Voice: 503-667-4517      Fax: 503-667-8863