matplotlib-users

Re: [Matplotlib-users] linearized log axis

[Christer <chm...@st...>]

Thank you for the help, I never knew what the symlog flag did actually.

However, there is still a slight problem:

=====================================================
x = array([0,1,2,4,6,9,12,24])
y = array([1000000, 500000, 100000, 100, 5, 1, 1, 1])
subplot(111)
plot(x, y)
yscale('symlog')
xscale=('linear')
ylim(-1,10000000)
show()
=====================================================

The plot looks exactly like I want it, the problem is when I change  
the "1"'s to "0"'s in the y-array, then I get a:

File "C:\Python26\lib\site-packages\matplotlib\ticker.py", line 1029,  
in is_decade
     lx = math.log(x)/math.log(base)
ValueError: math domain error

I suppose that means somewhere a log(0) is attempted. This kind of  
defeats the purpose...

/C

Quoting Eric Firing <ef...@ha...>:

> On 05/19/2010 10:28 AM, Benjamin Root wrote:
>> Maybe I am misunderstanding your problem, but you can select
'semilog'
>> for the x/yscale parameter.
>
> You mean "symlog".
>
> See
>
http://matplotlib.sourceforge.net/examples/pylab_examples/symlog_demo.html
>
> Although the example doesn't show it, the axis limits don't have to be
> symmetric.  For example, on the top plot, you can use
>
> gca().set_xlim([0, 100])
>
> to show only the right-hand side.
>
> Eric
>
>
>>
>> Ben Root
>>
>> On Wed, May 19, 2010 at 7:03 AM, Christer Malmberg
>> <Chr...@st...
>> <mailto:Chr...@st...>> wrote:
>>
>>     Hi,
>>
>>     my problem is that I need a graph with a discontinous y-axis. Let
me
>>     explain the problem: in my field (microbiology) the data
generated
>>     from for example growth assays have a huge range (10^0-10^9),
which
>>     has to be plotted on a semilogy style plot (cell concentration
vs.
>>     time). The problem is that 0 cells is a useful number to plot
>>     (indicates cell concentration lower than detection limit), but of
>>     course not possible to show in a log diagram. This is easily
solved on
>>     old-style logarithmic graph paper; since the data will be either
0, or
>>      >1 it is customary just to draw a zero x-axis at 10^-1 on the
paper
>>     and that's that. On the computer, this is extremely hard. Most
people
>>     I know resort to various tricks in Excel, such as entering a
small
>>     number (0.001 etc) and starting the y-axis range from 10^1 to
hide the
>>     problem. This makes excel draw a line, instead of leaving out the
dot
>>     and line entirely. The part of the curve below the x-axis is then
>>     manually cut off in a suitable image editor. Needless to say,
this is
>>     extremely kludgy. Even professional graphing packages like
Graphpad
>>     Prism resort to similar kludges (re-define 0 values to 0.1,
change the
>>     y-axis tick label to "0" etc.) This problem of course exists in
other
>>     fields, while investigating a solution I found a guy who worked
with
>>     aerosol contamination in clean rooms, and he needed to plot
values
>>     logarithmically, at the same time as showing detector noise
around
>>     1-10 particles. He solved it by the same trick I would like to do
in
>>     Matplotlib, namely plotting a standard semilogy plot but with the
>>     10^-1 to 10^0 decade being replaced by a 0-1 linear axis on the
same
>>     side.
>>
>>     The guy in this post has the same problem and a useful example:
>>     http://ubuntuforums.org/showthread.php?t=394851
>>
>>     His partial solution is quite bad though, and I just got stuck
while
>>     trying to improve it. I looked around the gallery for useful
examples,
>>     and the closest I could find is the twinx/twiny function, but I
didn't
>>     manage a plot that put one data curve across both axes.
>>
>>     This code gives an image that maybe explains what I'm trying to
do:
>>
>>     =======================================
>>     t = array([0,1,2,4,6,9,12,24])
>>     y = array([1000000, 500000, 100000, 100, 5, 1, 0, 0])
>>     subplot(111, xscale="linear", yscale="log")
>>     errorbar(x, y, yerr=0.4*y)
>>     linbit = axes([0.125, 0.1, 0.775, 0.1],frameon=False)
>>     linbit.xaxis.set_visible(False)
>>     for tl in linbit.get_yticklabels():
>>          tl.set_color('r')
>>     show()
>>     =======================================
>>
>>     (the y=0 points should be plotted and connected to the line in
the
>>     log part)
>>
>>     Is this possible to do in matplotlib? Could someone give me a
pointer
>>     on how to go on?
>>
>>     Sorry for the long mail,
>>
>>     /C
>>
>>
>>       
>>
------------------------------------------------------------------------------
>>
>>     _______________________________________________
>>     Matplotlib-users mailing list
>>     Mat...@li...
>>     <mailto:Mat...@li...>
>>     https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>>
>>
>>
>>
>>
------------------------------------------------------------------------------
>>
>>
>>
>>
>> _______________________________________________
>> Matplotlib-users mailing list
>> Mat...@li...
>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>
>
>
------------------------------------------------------------------------------
>
> _______________________________________________
> Matplotlib-users mailing list
> Mat...@li...
> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>

Re: [Matplotlib-users] linearized log axis

[Michael D. <md...@st...>]

Yep.  That's a bug.  Here's a patch to fix it:

ndex: lib/matplotlib/ticker.py
===================================================================
--- lib/matplotlib/ticker.py    (revision 8323)
+++ lib/matplotlib/ticker.py    (working copy)
@@ -1178,16 +1178,21 @@

  def decade_down(x, base=10):
      'floor x to the nearest lower decade'
-
+    if x == 0.0:
+        return -base
      lx = math.floor(math.log(x)/math.log(base))
      return base**lx

  def decade_up(x, base=10):
      'ceil x to the nearest higher decade'
+    if x == 0.0:
+        return base
      lx = math.ceil(math.log(x)/math.log(base))
      return base**lx

  def is_decade(x,base=10):
+    if x == 0.0:
+        return True
      lx = math.log(x)/math.log(base)
      return lx==int(lx)

Mike

On 05/20/2010 09:43 AM, Christer wrote:
> Thank you for the help, I never knew what the symlog flag did actually.
>
> However, there is still a slight problem:
>
> =====================================================
> x = array([0,1,2,4,6,9,12,24])
> y = array([1000000, 500000, 100000, 100, 5, 1, 1, 1])
> subplot(111)
> plot(x, y)
> yscale('symlog')
> xscale=('linear')
> ylim(-1,10000000)
> show()
> =====================================================
>
> The plot looks exactly like I want it, the problem is when I change
> the "1"'s to "0"'s in the y-array, then I get a:
>
> File "C:\Python26\lib\site-packages\matplotlib\ticker.py", line 1029,
> in is_decade
>       lx = math.log(x)/math.log(base)
> ValueError: math domain error
>
> I suppose that means somewhere a log(0) is attempted. This kind of
> defeats the purpose...
>
> /C
>
> Quoting Eric Firing<ef...@ha...>:
>
>    
>> On 05/19/2010 10:28 AM, Benjamin Root wrote:
>>      
>>> Maybe I am misunderstanding your problem, but you can select
>>>        
> 'semilog'
>    
>>> for the x/yscale parameter.
>>>        
>> You mean "symlog".
>>
>> See
>>
>>      
> http://matplotlib.sourceforge.net/examples/pylab_examples/symlog_demo.html
>    
>> Although the example doesn't show it, the axis limits don't have to be
>> symmetric.  For example, on the top plot, you can use
>>
>> gca().set_xlim([0, 100])
>>
>> to show only the right-hand side.
>>
>> Eric
>>
>>
>>      
>>> Ben Root
>>>
>>> On Wed, May 19, 2010 at 7:03 AM, Christer Malmberg
>>> <Chr...@st...
>>> <mailto:Chr...@st...>>  wrote:
>>>
>>>      Hi,
>>>
>>>      my problem is that I need a graph with a discontinous y-axis. Let
>>>        
> me
>    
>>>      explain the problem: in my field (microbiology) the data
>>>        
> generated
>    
>>>      from for example growth assays have a huge range (10^0-10^9),
>>>        
> which
>    
>>>      has to be plotted on a semilogy style plot (cell concentration
>>>        
> vs.
>    
>>>      time). The problem is that 0 cells is a useful number to plot
>>>      (indicates cell concentration lower than detection limit), but of
>>>      course not possible to show in a log diagram. This is easily
>>>        
> solved on
>    
>>>      old-style logarithmic graph paper; since the data will be either
>>>        
> 0, or
>    
>>>       >1 it is customary just to draw a zero x-axis at 10^-1 on the
>>>        
> paper
>    
>>>      and that's that. On the computer, this is extremely hard. Most
>>>        
> people
>    
>>>      I know resort to various tricks in Excel, such as entering a
>>>        
> small
>    
>>>      number (0.001 etc) and starting the y-axis range from 10^1 to
>>>        
> hide the
>    
>>>      problem. This makes excel draw a line, instead of leaving out the
>>>        
> dot
>    
>>>      and line entirely. The part of the curve below the x-axis is then
>>>      manually cut off in a suitable image editor. Needless to say,
>>>        
> this is
>    
>>>      extremely kludgy. Even professional graphing packages like
>>>        
> Graphpad
>    
>>>      Prism resort to similar kludges (re-define 0 values to 0.1,
>>>        
> change the
>    
>>>      y-axis tick label to "0" etc.) This problem of course exists in
>>>        
> other
>    
>>>      fields, while investigating a solution I found a guy who worked
>>>        
> with
>    
>>>      aerosol contamination in clean rooms, and he needed to plot
>>>        
> values
>    
>>>      logarithmically, at the same time as showing detector noise
>>>        
> around
>    
>>>      1-10 particles. He solved it by the same trick I would like to do
>>>        
> in
>    
>>>      Matplotlib, namely plotting a standard semilogy plot but with the
>>>      10^-1 to 10^0 decade being replaced by a 0-1 linear axis on the
>>>        
> same
>    
>>>      side.
>>>
>>>      The guy in this post has the same problem and a useful example:
>>>      http://ubuntuforums.org/showthread.php?t=394851
>>>
>>>      His partial solution is quite bad though, and I just got stuck
>>>        
> while
>    
>>>      trying to improve it. I looked around the gallery for useful
>>>        
> examples,
>    
>>>      and the closest I could find is the twinx/twiny function, but I
>>>        
> didn't
>    
>>>      manage a plot that put one data curve across both axes.
>>>
>>>      This code gives an image that maybe explains what I'm trying to
>>>        
> do:
>    
>>>      =======================================
>>>      t = array([0,1,2,4,6,9,12,24])
>>>      y = array([1000000, 500000, 100000, 100, 5, 1, 0, 0])
>>>      subplot(111, xscale="linear", yscale="log")
>>>      errorbar(x, y, yerr=0.4*y)
>>>      linbit = axes([0.125, 0.1, 0.775, 0.1],frameon=False)
>>>      linbit.xaxis.set_visible(False)
>>>      for tl in linbit.get_yticklabels():
>>>           tl.set_color('r')
>>>      show()
>>>      =======================================
>>>
>>>      (the y=0 points should be plotted and connected to the line in
>>>        
> the
>    
>>>      log part)
>>>
>>>      Is this possible to do in matplotlib? Could someone give me a
>>>        
> pointer
>    
>>>      on how to go on?
>>>
>>>      Sorry for the long mail,
>>>
>>>      /C
>>>
>>>
>>>
>>>
>>>        
> ------------------------------------------------------------------------------
>    
>>>      _______________________________________________
>>>      Matplotlib-users mailing list
>>>      Mat...@li...
>>>      <mailto:Mat...@li...>
>>>      https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>>>
>>>
>>>
>>>
>>>
>>>        
> ------------------------------------------------------------------------------
>    
>>>
>>>
>>>
>>> _______________________________________________
>>> Matplotlib-users mailing list
>>> Mat...@li...
>>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>>>        
>>
>>
>>      
> ------------------------------------------------------------------------------
>    
>> _______________________________________________
>> Matplotlib-users mailing list
>> Mat...@li...
>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>>
>>      
>
>
> ------------------------------------------------------------------------------
>
> _______________________________________________
> Matplotlib-users mailing list
> Mat...@li...
> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>    


-- 
Michael Droettboom
Science Software Branch
Space Telescope Science Institute
Baltimore, Maryland, USA

Re: [Matplotlib-users] linearized log axis

[Benjamin R. <ben...@ou...>]

Do we really want to depend on a floating point equality?

Ben Root

On Thu, May 20, 2010 at 9:02 AM, Michael Droettboom <md...@st...> wrote:

> Yep.  That's a bug.  Here's a patch to fix it:
>
> ndex: lib/matplotlib/ticker.py
> ===================================================================
> --- lib/matplotlib/ticker.py    (revision 8323)
> +++ lib/matplotlib/ticker.py    (working copy)
> @@ -1178,16 +1178,21 @@
>
>  def decade_down(x, base=10):
>      'floor x to the nearest lower decade'
> -
> +    if x == 0.0:
> +        return -base
>      lx = math.floor(math.log(x)/math.log(base))
>      return base**lx
>
>  def decade_up(x, base=10):
>      'ceil x to the nearest higher decade'
> +    if x == 0.0:
> +        return base
>      lx = math.ceil(math.log(x)/math.log(base))
>      return base**lx
>
>  def is_decade(x,base=10):
> +    if x == 0.0:
> +        return True
>       lx = math.log(x)/math.log(base)
>       return lx==int(lx)
>
> Mike
>
> On 05/20/2010 09:43 AM, Christer wrote:
> > Thank you for the help, I never knew what the symlog flag did actually.
> >
> > However, there is still a slight problem:
> >
> > =====================================================
> > x = array([0,1,2,4,6,9,12,24])
> > y = array([1000000, 500000, 100000, 100, 5, 1, 1, 1])
> > subplot(111)
> > plot(x, y)
> > yscale('symlog')
> > xscale=('linear')
> > ylim(-1,10000000)
> > show()
> > =====================================================
> >
> > The plot looks exactly like I want it, the problem is when I change
> > the "1"'s to "0"'s in the y-array, then I get a:
> >
> > File "C:\Python26\lib\site-packages\matplotlib\ticker.py", line 1029,
> > in is_decade
> >       lx = math.log(x)/math.log(base)
> > ValueError: math domain error
> >
> > I suppose that means somewhere a log(0) is attempted. This kind of
> > defeats the purpose...
> >
> > /C
> >
> > Quoting Eric Firing<ef...@ha...>:
> >
> >
> >> On 05/19/2010 10:28 AM, Benjamin Root wrote:
> >>
> >>> Maybe I am misunderstanding your problem, but you can select
> >>>
> > 'semilog'
> >
> >>> for the x/yscale parameter.
> >>>
> >> You mean "symlog".
> >>
> >> See
> >>
> >>
> >
> http://matplotlib.sourceforge.net/examples/pylab_examples/symlog_demo.html
> >
> >> Although the example doesn't show it, the axis limits don't have to be
> >> symmetric.  For example, on the top plot, you can use
> >>
> >> gca().set_xlim([0, 100])
> >>
> >> to show only the right-hand side.
> >>
> >> Eric
> >>
> >>
> >>
> >>> Ben Root
> >>>
> >>> On Wed, May 19, 2010 at 7:03 AM, Christer Malmberg
> >>> <Chr...@st...
> >>> <mailto:Chr...@st...>>  wrote:
> >>>
> >>>      Hi,
> >>>
> >>>      my problem is that I need a graph with a discontinous y-axis. Let
> >>>
> > me
> >
> >>>      explain the problem: in my field (microbiology) the data
> >>>
> > generated
> >
> >>>      from for example growth assays have a huge range (10^0-10^9),
> >>>
> > which
> >
> >>>      has to be plotted on a semilogy style plot (cell concentration
> >>>
> > vs.
> >
> >>>      time). The problem is that 0 cells is a useful number to plot
> >>>      (indicates cell concentration lower than detection limit), but of
> >>>      course not possible to show in a log diagram. This is easily
> >>>
> > solved on
> >
> >>>      old-style logarithmic graph paper; since the data will be either
> >>>
> > 0, or
> >
> >>>       >1 it is customary just to draw a zero x-axis at 10^-1 on the
> >>>
> > paper
> >
> >>>      and that's that. On the computer, this is extremely hard. Most
> >>>
> > people
> >
> >>>      I know resort to various tricks in Excel, such as entering a
> >>>
> > small
> >
> >>>      number (0.001 etc) and starting the y-axis range from 10^1 to
> >>>
> > hide the
> >
> >>>      problem. This makes excel draw a line, instead of leaving out the
> >>>
> > dot
> >
> >>>      and line entirely. The part of the curve below the x-axis is then
> >>>      manually cut off in a suitable image editor. Needless to say,
> >>>
> > this is
> >
> >>>      extremely kludgy. Even professional graphing packages like
> >>>
> > Graphpad
> >
> >>>      Prism resort to similar kludges (re-define 0 values to 0.1,
> >>>
> > change the
> >
> >>>      y-axis tick label to "0" etc.) This problem of course exists in
> >>>
> > other
> >
> >>>      fields, while investigating a solution I found a guy who worked
> >>>
> > with
> >
> >>>      aerosol contamination in clean rooms, and he needed to plot
> >>>
> > values
> >
> >>>      logarithmically, at the same time as showing detector noise
> >>>
> > around
> >
> >>>      1-10 particles. He solved it by the same trick I would like to do
> >>>
> > in
> >
> >>>      Matplotlib, namely plotting a standard semilogy plot but with the
> >>>      10^-1 to 10^0 decade being replaced by a 0-1 linear axis on the
> >>>
> > same
> >
> >>>      side.
> >>>
> >>>      The guy in this post has the same problem and a useful example:
> >>>      http://ubuntuforums.org/showthread.php?t=394851
> >>>
> >>>      His partial solution is quite bad though, and I just got stuck
> >>>
> > while
> >
> >>>      trying to improve it. I looked around the gallery for useful
> >>>
> > examples,
> >
> >>>      and the closest I could find is the twinx/twiny function, but I
> >>>
> > didn't
> >
> >>>      manage a plot that put one data curve across both axes.
> >>>
> >>>      This code gives an image that maybe explains what I'm trying to
> >>>
> > do:
> >
> >>>      =======================================
> >>>      t = array([0,1,2,4,6,9,12,24])
> >>>      y = array([1000000, 500000, 100000, 100, 5, 1, 0, 0])
> >>>      subplot(111, xscale="linear", yscale="log")
> >>>      errorbar(x, y, yerr=0.4*y)
> >>>      linbit = axes([0.125, 0.1, 0.775, 0.1],frameon=False)
> >>>      linbit.xaxis.set_visible(False)
> >>>      for tl in linbit.get_yticklabels():
> >>>           tl.set_color('r')
> >>>      show()
> >>>      =======================================
> >>>
> >>>      (the y=0 points should be plotted and connected to the line in
> >>>
> > the
> >
> >>>      log part)
> >>>
> >>>      Is this possible to do in matplotlib? Could someone give me a
> >>>
> > pointer
> >
> >>>      on how to go on?
> >>>
> >>>      Sorry for the long mail,
> >>>
> >>>      /C
> >>>
> >>>
> >>>
> >>>
> >>>
> >
> ------------------------------------------------------------------------------
> >
> >>>      _______________________________________________
> >>>      Matplotlib-users mailing list
> >>>      Mat...@li...
> >>>      <mailto:Mat...@li...>
> >>>      https://lists.sourceforge.net/lists/listinfo/matplotlib-users
> >>>
> >>>
> >>>
> >>>
> >>>
> >>>
> >
> ------------------------------------------------------------------------------
> >
> >>>
> >>>
> >>>
> >>> _______________________________________________
> >>> Matplotlib-users mailing list
> >>> Mat...@li...
> >>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
> >>>
> >>
> >>
> >>
> >
> ------------------------------------------------------------------------------
> >
> >> _______________________________________________
> >> Matplotlib-users mailing list
> >> Mat...@li...
> >> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
> >>
> >>
> >
> >
> >
> ------------------------------------------------------------------------------
> >
> > _______________________________________________
> > Matplotlib-users mailing list
> > Mat...@li...
> > https://lists.sourceforge.net/lists/listinfo/matplotlib-users
> >
>
>
> --
> Michael Droettboom
> Science Software Branch
> Space Telescope Science Institute
> Baltimore, Maryland, USA
>
>
>
> ------------------------------------------------------------------------------
>
> _______________________________________________
> Matplotlib-users mailing list
> Mat...@li...
> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>

Re: [Matplotlib-users] linearized log axis

[Michael D. <md...@st...>]

In this case, yes.  The assumption of these (private) functions is that 
x will be non-negative.  The only case where we need to worry about log 
raising an exception is with exactly 0.

Mike

On 05/20/2010 10:08 AM, Benjamin Root wrote:
> Do we really want to depend on a floating point equality?
>
> Ben Root
>
> On Thu, May 20, 2010 at 9:02 AM, Michael Droettboom<md...@st...>  wrote:
>
>    
>> Yep.  That's a bug.  Here's a patch to fix it:
>>
>> ndex: lib/matplotlib/ticker.py
>> ===================================================================
>> --- lib/matplotlib/ticker.py    (revision 8323)
>> +++ lib/matplotlib/ticker.py    (working copy)
>> @@ -1178,16 +1178,21 @@
>>
>>   def decade_down(x, base=10):
>>       'floor x to the nearest lower decade'
>> -
>> +    if x == 0.0:
>> +        return -base
>>       lx = math.floor(math.log(x)/math.log(base))
>>       return base**lx
>>
>>   def decade_up(x, base=10):
>>       'ceil x to the nearest higher decade'
>> +    if x == 0.0:
>> +        return base
>>       lx = math.ceil(math.log(x)/math.log(base))
>>       return base**lx
>>
>>   def is_decade(x,base=10):
>> +    if x == 0.0:
>> +        return True
>>        lx = math.log(x)/math.log(base)
>>        return lx==int(lx)
>>
>> Mike
>>
>> On 05/20/2010 09:43 AM, Christer wrote:
>>      
>>> Thank you for the help, I never knew what the symlog flag did actually.
>>>
>>> However, there is still a slight problem:
>>>
>>> =====================================================
>>> x = array([0,1,2,4,6,9,12,24])
>>> y = array([1000000, 500000, 100000, 100, 5, 1, 1, 1])
>>> subplot(111)
>>> plot(x, y)
>>> yscale('symlog')
>>> xscale=('linear')
>>> ylim(-1,10000000)
>>> show()
>>> =====================================================
>>>
>>> The plot looks exactly like I want it, the problem is when I change
>>> the "1"'s to "0"'s in the y-array, then I get a:
>>>
>>> File "C:\Python26\lib\site-packages\matplotlib\ticker.py", line 1029,
>>> in is_decade
>>>        lx = math.log(x)/math.log(base)
>>> ValueError: math domain error
>>>
>>> I suppose that means somewhere a log(0) is attempted. This kind of
>>> defeats the purpose...
>>>
>>> /C
>>>
>>> Quoting Eric Firing<ef...@ha...>:
>>>
>>>
>>>        
>>>> On 05/19/2010 10:28 AM, Benjamin Root wrote:
>>>>
>>>>          
>>>>> Maybe I am misunderstanding your problem, but you can select
>>>>>
>>>>>            
>>> 'semilog'
>>>
>>>        
>>>>> for the x/yscale parameter.
>>>>>
>>>>>            
>>>> You mean "symlog".
>>>>
>>>> See
>>>>
>>>>
>>>>          
>>>        
>> http://matplotlib.sourceforge.net/examples/pylab_examples/symlog_demo.html
>>      
>>>        
>>>> Although the example doesn't show it, the axis limits don't have to be
>>>> symmetric.  For example, on the top plot, you can use
>>>>
>>>> gca().set_xlim([0, 100])
>>>>
>>>> to show only the right-hand side.
>>>>
>>>> Eric
>>>>
>>>>
>>>>
>>>>          
>>>>> Ben Root
>>>>>
>>>>> On Wed, May 19, 2010 at 7:03 AM, Christer Malmberg
>>>>> <Chr...@st...
>>>>> <mailto:Chr...@st...>>   wrote:
>>>>>
>>>>>       Hi,
>>>>>
>>>>>       my problem is that I need a graph with a discontinous y-axis. Let
>>>>>
>>>>>            
>>> me
>>>
>>>        
>>>>>       explain the problem: in my field (microbiology) the data
>>>>>
>>>>>            
>>> generated
>>>
>>>        
>>>>>       from for example growth assays have a huge range (10^0-10^9),
>>>>>
>>>>>            
>>> which
>>>
>>>        
>>>>>       has to be plotted on a semilogy style plot (cell concentration
>>>>>
>>>>>            
>>> vs.
>>>
>>>        
>>>>>       time). The problem is that 0 cells is a useful number to plot
>>>>>       (indicates cell concentration lower than detection limit), but of
>>>>>       course not possible to show in a log diagram. This is easily
>>>>>
>>>>>            
>>> solved on
>>>
>>>        
>>>>>       old-style logarithmic graph paper; since the data will be either
>>>>>
>>>>>            
>>> 0, or
>>>
>>>        
>>>>>        >1 it is customary just to draw a zero x-axis at 10^-1 on the
>>>>>
>>>>>            
>>> paper
>>>
>>>        
>>>>>       and that's that. On the computer, this is extremely hard. Most
>>>>>
>>>>>            
>>> people
>>>
>>>        
>>>>>       I know resort to various tricks in Excel, such as entering a
>>>>>
>>>>>            
>>> small
>>>
>>>        
>>>>>       number (0.001 etc) and starting the y-axis range from 10^1 to
>>>>>
>>>>>            
>>> hide the
>>>
>>>        
>>>>>       problem. This makes excel draw a line, instead of leaving out the
>>>>>
>>>>>            
>>> dot
>>>
>>>        
>>>>>       and line entirely. The part of the curve below the x-axis is then
>>>>>       manually cut off in a suitable image editor. Needless to say,
>>>>>
>>>>>            
>>> this is
>>>
>>>        
>>>>>       extremely kludgy. Even professional graphing packages like
>>>>>
>>>>>            
>>> Graphpad
>>>
>>>        
>>>>>       Prism resort to similar kludges (re-define 0 values to 0.1,
>>>>>
>>>>>            
>>> change the
>>>
>>>        
>>>>>       y-axis tick label to "0" etc.) This problem of course exists in
>>>>>
>>>>>            
>>> other
>>>
>>>        
>>>>>       fields, while investigating a solution I found a guy who worked
>>>>>
>>>>>            
>>> with
>>>
>>>        
>>>>>       aerosol contamination in clean rooms, and he needed to plot
>>>>>
>>>>>            
>>> values
>>>
>>>        
>>>>>       logarithmically, at the same time as showing detector noise
>>>>>
>>>>>            
>>> around
>>>
>>>        
>>>>>       1-10 particles. He solved it by the same trick I would like to do
>>>>>
>>>>>            
>>> in
>>>
>>>        
>>>>>       Matplotlib, namely plotting a standard semilogy plot but with the
>>>>>       10^-1 to 10^0 decade being replaced by a 0-1 linear axis on the
>>>>>
>>>>>            
>>> same
>>>
>>>        
>>>>>       side.
>>>>>
>>>>>       The guy in this post has the same problem and a useful example:
>>>>>       http://ubuntuforums.org/showthread.php?t=394851
>>>>>
>>>>>       His partial solution is quite bad though, and I just got stuck
>>>>>
>>>>>            
>>> while
>>>
>>>        
>>>>>       trying to improve it. I looked around the gallery for useful
>>>>>
>>>>>            
>>> examples,
>>>
>>>        
>>>>>       and the closest I could find is the twinx/twiny function, but I
>>>>>
>>>>>            
>>> didn't
>>>
>>>        
>>>>>       manage a plot that put one data curve across both axes.
>>>>>
>>>>>       This code gives an image that maybe explains what I'm trying to
>>>>>
>>>>>            
>>> do:
>>>
>>>        
>>>>>       =======================================
>>>>>       t = array([0,1,2,4,6,9,12,24])
>>>>>       y = array([1000000, 500000, 100000, 100, 5, 1, 0, 0])
>>>>>       subplot(111, xscale="linear", yscale="log")
>>>>>       errorbar(x, y, yerr=0.4*y)
>>>>>       linbit = axes([0.125, 0.1, 0.775, 0.1],frameon=False)
>>>>>       linbit.xaxis.set_visible(False)
>>>>>       for tl in linbit.get_yticklabels():
>>>>>            tl.set_color('r')
>>>>>       show()
>>>>>       =======================================
>>>>>
>>>>>       (the y=0 points should be plotted and connected to the line in
>>>>>
>>>>>            
>>> the
>>>
>>>        
>>>>>       log part)
>>>>>
>>>>>       Is this possible to do in matplotlib? Could someone give me a
>>>>>
>>>>>            
>>> pointer
>>>
>>>        
>>>>>       on how to go on?
>>>>>
>>>>>       Sorry for the long mail,
>>>>>
>>>>>       /C
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>            
>>>        
>> ------------------------------------------------------------------------------
>>      
>>>        
>>>>>       _______________________________________________
>>>>>       Matplotlib-users mailing list
>>>>>       Mat...@li...
>>>>>       <mailto:Mat...@li...>
>>>>>       https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>            
>>>        
>> ------------------------------------------------------------------------------
>>      
>>>        
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Matplotlib-users mailing list
>>>>> Mat...@li...
>>>>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>>>>>
>>>>>            
>>>>
>>>>
>>>>          
>>>        
>> ------------------------------------------------------------------------------
>>      
>>>        
>>>> _______________________________________________
>>>> Matplotlib-users mailing list
>>>> Mat...@li...
>>>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>>>>
>>>>
>>>>          
>>>
>>>
>>>        
>> ------------------------------------------------------------------------------
>>      
>>> _______________________________________________
>>> Matplotlib-users mailing list
>>> Mat...@li...
>>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>>>
>>>        
>>
>> --
>> Michael Droettboom
>> Science Software Branch
>> Space Telescope Science Institute
>> Baltimore, Maryland, USA
>>
>>
>>
>> ------------------------------------------------------------------------------
>>
>> _______________________________________________
>> Matplotlib-users mailing list
>> Mat...@li...
>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>>
>>      
>    


-- 
Michael Droettboom
Science Software Branch
Space Telescope Science Institute
Baltimore, Maryland, USA

Re: [Matplotlib-users] linearized log axis

[Benjamin R. <ben...@ou...>]

Ok, good, I just wanted to do a sanity check.

On Thu, May 20, 2010 at 9:21 AM, Michael Droettboom <md...@st...> wrote:

> In this case, yes.  The assumption of these (private) functions is that
> x will be non-negative.  The only case where we need to worry about log
> raising an exception is with exactly 0.
>
> Mike
>
> On 05/20/2010 10:08 AM, Benjamin Root wrote:
> > Do we really want to depend on a floating point equality?
> >
> > Ben Root
> >
> > On Thu, May 20, 2010 at 9:02 AM, Michael Droettboom<md...@st...>
>  wrote:
> >
> >
> >> Yep.  That's a bug.  Here's a patch to fix it:
> >>
> >> ndex: lib/matplotlib/ticker.py
> >> ===================================================================
> >> --- lib/matplotlib/ticker.py    (revision 8323)
> >> +++ lib/matplotlib/ticker.py    (working copy)
> >> @@ -1178,16 +1178,21 @@
> >>
> >>   def decade_down(x, base=10):
> >>       'floor x to the nearest lower decade'
> >> -
> >> +    if x == 0.0:
> >> +        return -base
> >>       lx = math.floor(math.log(x)/math.log(base))
> >>       return base**lx
> >>
> >>   def decade_up(x, base=10):
> >>       'ceil x to the nearest higher decade'
> >> +    if x == 0.0:
> >> +        return base
> >>       lx = math.ceil(math.log(x)/math.log(base))
> >>       return base**lx
> >>
> >>   def is_decade(x,base=10):
> >> +    if x == 0.0:
> >> +        return True
> >>        lx = math.log(x)/math.log(base)
> >>        return lx==int(lx)
> >>
> >> Mike
> >>
> >> On 05/20/2010 09:43 AM, Christer wrote:
> >>
> >>> Thank you for the help, I never knew what the symlog flag did actually.
> >>>
> >>> However, there is still a slight problem:
> >>>
> >>> =====================================================
> >>> x = array([0,1,2,4,6,9,12,24])
> >>> y = array([1000000, 500000, 100000, 100, 5, 1, 1, 1])
> >>> subplot(111)
> >>> plot(x, y)
> >>> yscale('symlog')
> >>> xscale=('linear')
> >>> ylim(-1,10000000)
> >>> show()
> >>> =====================================================
> >>>
> >>> The plot looks exactly like I want it, the problem is when I change
> >>> the "1"'s to "0"'s in the y-array, then I get a:
> >>>
> >>> File "C:\Python26\lib\site-packages\matplotlib\ticker.py", line 1029,
> >>> in is_decade
> >>>        lx = math.log(x)/math.log(base)
> >>> ValueError: math domain error
> >>>
> >>> I suppose that means somewhere a log(0) is attempted. This kind of
> >>> defeats the purpose...
> >>>
> >>> /C
> >>>
> >>> Quoting Eric Firing<ef...@ha...>:
> >>>
> >>>
> >>>
> >>>> On 05/19/2010 10:28 AM, Benjamin Root wrote:
> >>>>
> >>>>
> >>>>> Maybe I am misunderstanding your problem, but you can select
> >>>>>
> >>>>>
> >>> 'semilog'
> >>>
> >>>
> >>>>> for the x/yscale parameter.
> >>>>>
> >>>>>
> >>>> You mean "symlog".
> >>>>
> >>>> See
> >>>>
> >>>>
> >>>>
> >>>
> >>
> http://matplotlib.sourceforge.net/examples/pylab_examples/symlog_demo.html
> >>
> >>>
> >>>> Although the example doesn't show it, the axis limits don't have to be
> >>>> symmetric.  For example, on the top plot, you can use
> >>>>
> >>>> gca().set_xlim([0, 100])
> >>>>
> >>>> to show only the right-hand side.
> >>>>
> >>>> Eric
> >>>>
> >>>>
> >>>>
> >>>>
> >>>>> Ben Root
> >>>>>
> >>>>> On Wed, May 19, 2010 at 7:03 AM, Christer Malmberg
> >>>>> <Chr...@st...
> >>>>> <mailto:Chr...@st...>>   wrote:
> >>>>>
> >>>>>       Hi,
> >>>>>
> >>>>>       my problem is that I need a graph with a discontinous y-axis.
> Let
> >>>>>
> >>>>>
> >>> me
> >>>
> >>>
> >>>>>       explain the problem: in my field (microbiology) the data
> >>>>>
> >>>>>
> >>> generated
> >>>
> >>>
> >>>>>       from for example growth assays have a huge range (10^0-10^9),
> >>>>>
> >>>>>
> >>> which
> >>>
> >>>
> >>>>>       has to be plotted on a semilogy style plot (cell concentration
> >>>>>
> >>>>>
> >>> vs.
> >>>
> >>>
> >>>>>       time). The problem is that 0 cells is a useful number to plot
> >>>>>       (indicates cell concentration lower than detection limit), but
> of
> >>>>>       course not possible to show in a log diagram. This is easily
> >>>>>
> >>>>>
> >>> solved on
> >>>
> >>>
> >>>>>       old-style logarithmic graph paper; since the data will be
> either
> >>>>>
> >>>>>
> >>> 0, or
> >>>
> >>>
> >>>>>        >1 it is customary just to draw a zero x-axis at 10^-1 on the
> >>>>>
> >>>>>
> >>> paper
> >>>
> >>>
> >>>>>       and that's that. On the computer, this is extremely hard. Most
> >>>>>
> >>>>>
> >>> people
> >>>
> >>>
> >>>>>       I know resort to various tricks in Excel, such as entering a
> >>>>>
> >>>>>
> >>> small
> >>>
> >>>
> >>>>>       number (0.001 etc) and starting the y-axis range from 10^1 to
> >>>>>
> >>>>>
> >>> hide the
> >>>
> >>>
> >>>>>       problem. This makes excel draw a line, instead of leaving out
> the
> >>>>>
> >>>>>
> >>> dot
> >>>
> >>>
> >>>>>       and line entirely. The part of the curve below the x-axis is
> then
> >>>>>       manually cut off in a suitable image editor. Needless to say,
> >>>>>
> >>>>>
> >>> this is
> >>>
> >>>
> >>>>>       extremely kludgy. Even professional graphing packages like
> >>>>>
> >>>>>
> >>> Graphpad
> >>>
> >>>
> >>>>>       Prism resort to similar kludges (re-define 0 values to 0.1,
> >>>>>
> >>>>>
> >>> change the
> >>>
> >>>
> >>>>>       y-axis tick label to "0" etc.) This problem of course exists in
> >>>>>
> >>>>>
> >>> other
> >>>
> >>>
> >>>>>       fields, while investigating a solution I found a guy who worked
> >>>>>
> >>>>>
> >>> with
> >>>
> >>>
> >>>>>       aerosol contamination in clean rooms, and he needed to plot
> >>>>>
> >>>>>
> >>> values
> >>>
> >>>
> >>>>>       logarithmically, at the same time as showing detector noise
> >>>>>
> >>>>>
> >>> around
> >>>
> >>>
> >>>>>       1-10 particles. He solved it by the same trick I would like to
> do
> >>>>>
> >>>>>
> >>> in
> >>>
> >>>
> >>>>>       Matplotlib, namely plotting a standard semilogy plot but with
> the
> >>>>>       10^-1 to 10^0 decade being replaced by a 0-1 linear axis on the
> >>>>>
> >>>>>
> >>> same
> >>>
> >>>
> >>>>>       side.
> >>>>>
> >>>>>       The guy in this post has the same problem and a useful example:
> >>>>>       http://ubuntuforums.org/showthread.php?t=394851
> >>>>>
> >>>>>       His partial solution is quite bad though, and I just got stuck
> >>>>>
> >>>>>
> >>> while
> >>>
> >>>
> >>>>>       trying to improve it. I looked around the gallery for useful
> >>>>>
> >>>>>
> >>> examples,
> >>>
> >>>
> >>>>>       and the closest I could find is the twinx/twiny function, but I
> >>>>>
> >>>>>
> >>> didn't
> >>>
> >>>
> >>>>>       manage a plot that put one data curve across both axes.
> >>>>>
> >>>>>       This code gives an image that maybe explains what I'm trying to
> >>>>>
> >>>>>
> >>> do:
> >>>
> >>>
> >>>>>       =======================================
> >>>>>       t = array([0,1,2,4,6,9,12,24])
> >>>>>       y = array([1000000, 500000, 100000, 100, 5, 1, 0, 0])
> >>>>>       subplot(111, xscale="linear", yscale="log")
> >>>>>       errorbar(x, y, yerr=0.4*y)
> >>>>>       linbit = axes([0.125, 0.1, 0.775, 0.1],frameon=False)
> >>>>>       linbit.xaxis.set_visible(False)
> >>>>>       for tl in linbit.get_yticklabels():
> >>>>>            tl.set_color('r')
> >>>>>       show()
> >>>>>       =======================================
> >>>>>
> >>>>>       (the y=0 points should be plotted and connected to the line in
> >>>>>
> >>>>>
> >>> the
> >>>
> >>>
> >>>>>       log part)
> >>>>>
> >>>>>       Is this possible to do in matplotlib? Could someone give me a
> >>>>>
> >>>>>
> >>> pointer
> >>>
> >>>
> >>>>>       on how to go on?
> >>>>>
> >>>>>       Sorry for the long mail,
> >>>>>
> >>>>>       /C
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>
> >>
> ------------------------------------------------------------------------------
> >>
> >>>
> >>>>>       _______________________________________________
> >>>>>       Matplotlib-users mailing list
> >>>>>       Mat...@li...
> >>>>>       <mailto:Mat...@li...>
> >>>>>       https://lists.sourceforge.net/lists/listinfo/matplotlib-users
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>
> >>
> ------------------------------------------------------------------------------
> >>
> >>>
> >>>>>
> >>>>>
> >>>>> _______________________________________________
> >>>>> Matplotlib-users mailing list
> >>>>> Mat...@li...
> >>>>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
> >>>>>
> >>>>>
> >>>>
> >>>>
> >>>>
> >>>
> >>
> ------------------------------------------------------------------------------
> >>
> >>>
> >>>> _______________________________________________
> >>>> Matplotlib-users mailing list
> >>>> Mat...@li...
> >>>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
> >>>>
> >>>>
> >>>>
> >>>
> >>>
> >>>
> >>
> ------------------------------------------------------------------------------
> >>
> >>> _______________________________________________
> >>> Matplotlib-users mailing list
> >>> Mat...@li...
> >>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
> >>>
> >>>
> >>
> >> --
> >> Michael Droettboom
> >> Science Software Branch
> >> Space Telescope Science Institute
> >> Baltimore, Maryland, USA
> >>
> >>
> >>
> >>
> ------------------------------------------------------------------------------
> >>
> >> _______________________________________________
> >> Matplotlib-users mailing list
> >> Mat...@li...
> >> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
> >>
> >>
> >
>
>
> --
> Michael Droettboom
> Science Software Branch
> Space Telescope Science Institute
> Baltimore, Maryland, USA
>
>
>
> ------------------------------------------------------------------------------
>
> _______________________________________________
> Matplotlib-users mailing list
> Mat...@li...
> https://lists.sourceforge.net/lists/listinfo/matplotlib-users
>