Re: [PyX-user] PyX with scipy/numpy

[Andre W. <wo...@us...>]

Hi all,

On 25.07.07, BK wrote:
> > is there any way to directly plot scipy/numpy-type arrays? I am reading
> my data using the scipy.array_import.read_array() method and then doing
> some data manipulations based on the scipy arrays.
> Before plotting I currently write the data from the arrays into a list
> which can be read by the pyx.graph.graphxy.plot() method. Would it be
> possible to add scipy-type array data to the plot() method of a graph?

Sure ... and its even trivial to implement new data providers for pyx.
(Let me call instances of the classes defined in pyx.graph.data to be
"data providers".) Well, there are some non-trivial features in the
data providers. Think of functions, where the range the function is
evaluated may depend on the range of the corresponding axes. But those
things really are advanced features and simple data providers don't
need to deal with such stuff at all.

The point is that (1) I do not know what people want and need and (2)
I'm not an scipy user at all. OTOH I do know that pyx.graph.data
really is very limited in terms of manipulating data. And it's also
slow. So we just need to find out what's needed for all of you, guys,
and I'll try to add an appropriate data provider to PyX. I'm sure
thats better than reimplementing features found in scipy. It's fine
that PyX comes with some useful data providers, since PyX users really
should not need to scipy in the first place. But those of you who are
beeing happy scipy users should get the power out of scipy + PyX most
easily. Ok, enough blah, blah ...

There is the pyx.graph.data.list class, which takes a list of lists
and an columns dictionary. For example:

d = pyx.graph.data.list([[1, 2], [3, 4]], x=1, y=2)

(and than proceed with g.plot(d) ...)

This means that you have two points, the first one with x=1 and y=2
and the second one with x=3 and y=4.

I've just tried to use read_array and provide a useful "adapter" for
PyX. This looks the following:

    from scipy.io import read_array
    from pyx import *

    class lists(graph.data._data):

        defaultstyles = [graph.style.symbol()]

        def __init__(self, title="user provided lists", **columns):
            self.columns = columns
            self.columnnames = columns.keys()
            self.title = title

    d = read_array("scipytest.dat")
    g = graph.graphxy(width=10)
    g.plot(lists(x=d[0], y=d[1]))
    g.writeEPSfile("scipytest")

It's that simple. You really need to understand, that the PyX data
instances are very thin wrappers and maybe we should even more focus
on that. While I kind of dislike the above notation (d[0] and d[1] are
completely independed) and thus I focused on "points" in
pyx.graph.data.list, why not just add the lists class to PyX as well?
Maybe the name is not perfect. Well, "list" was not a good idea at
all ... I could think of renaming the old "list" to "points" and
adding the "lists" class as shown above. I know that'll break quite
some code and we could add some deprecated "list = points" to
pyx.graph.data, but let's keep the focus: Does the signature of the
lists class shown above match the scipy users needs? Or should it look
differently?


André

-- 
by  _ _      _    Dr. André Wobst
   / \ \    / )   wo...@us..., http://www.wobsta.de/
  / _ \ \/\/ /    PyX - High quality PostScript and PDF figures
 (_/ \_)_/\_/     with Python & TeX: visit http://pyx.sourceforge.net/