error when plotting log'd array in matplotlib/scipy/numpy

Question

I have two arrays and I take their logs. When I do that and try to plot their scatter plot, I get this error:

  File "/Library/Python/2.6/site-packages/matplotlib-1.0.svn_r7892-py2.6-macosx-10.6-universal.egg/matplotlib/pyplot.py", line 2192, in scatter
    ret = ax.scatter(x, y, s, c, marker, cmap, norm, vmin, vmax, alpha, linewidths, faceted, verts, **kwargs)
  File "/Library/Python/2.6/site-packages/matplotlib-1.0.svn_r7892-py2.6-macosx-10.6-universal.egg/matplotlib/axes.py", line 5384, in scatter
    self.add_collection(collection)
  File "/Library/Python/2.6/site-packages/matplotlib-1.0.svn_r7892-py2.6-macosx-10.6-universal.egg/matplotlib/axes.py", line 1391, in add_collection
    self.update_datalim(collection.get_datalim(self.transData))
  File "/Library/Python/2.6/site-packages/matplotlib-1.0.svn_r7892-py2.6-macosx-10.6-universal.egg/matplotlib/collections.py", line 153, in get_datalim
    offsets = transOffset.transform_non_affine(offsets)
  File "/Library/Python/2.6/site-packages/matplotlib-1.0.svn_r7892-py2.6-macosx-10.6-universal.egg/matplotlib/transforms.py", line 1924, in transform_non_affine
    self._a.transform(points))
 File "/Library/Python/2.6/site-packages/matplotlib-1.0.svn_r7892-py2.6-macosx-10.6-universal.egg/matplotlib/transforms.py", line 1420, in transform
    return affine_transform(points, mtx)
ValueError: Invalid vertices array.

the code is simply:

myarray_x = log(my_array[:, 0])
myarray_y = log(my_array[:, 1])
plt.scatter(myarray_x, myarray_y)

any idea what could be causing this? thanks.

Answer 1

This runs successfully for me

>>> from numpy import log, array
>>> import matplotlib.pyplot as plt
>>> my_array = array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]])
>>> my_array
array([[ 1.,  2.],
       [ 3.,  4.],
       [ 5.,  6.]])
>>> myarray_x = log(my_array[:, 0])
>>> myarray_y = log(my_array[:, 1])
>>> plt.scatter(myarray_x, myarray_y)
<matplotlib.collections.CircleCollection object at 0x030C7A10>
>>> plt.show()

so perhaps the problem is with something you haven't shown us. Can you provide a complete piece of sample code exactly as you ran it so we can reproduce your problem?

Answer 2

New Answer:

From looking at the source, this error is thrown if the points array passed into affine_transform is of the wrong dimensions or if it is empty. Here are the relevant lines:

if (!vertices ||
        (PyArray_NDIM(vertices) == 2 && PyArray_DIM(vertices, 1) != 2) ||
        (PyArray_NDIM(vertices) == 1 && PyArray_DIM(vertices, 0) != 2))
         throw Py::ValueError("Invalid vertices array.");

Take a look at the dimensions of your myarray_x and myarray_y arrays before going ahead maybe.

Old Answer:

My best guess it that you are taking the log of values <= 0. This will either give you nan's or -infs(in the case of being equal to 0) in your arrays which of course it can't plot.

(wiso is right though - these points are just ignored)

Answer 3

I had the same problem which I fixed recently:

The problem for me was that my X and Y (numpy) arrays were made up of 128 bit floats.

The solution in this case was to recast the arrays to a lower precision float i.e.

array = numpy.float64(array)

Hope this helps :~)