Hello,

I've run into a strange behaviour of matplotlib while trying to figure

out why my data was displayed incorrectly.

I'm note quite sure if this is a bug or expected behaviour, but I feel

it's kind of counter-intuitive, so I'm posting here.

The attached python script does a scatter plot of some data. I'm using

the first column as the x coordinates and the second as

y. Looking at the matrix (x == y), I'd expect the three data point to

be on a diagonal line.

Now there seem to be a difference on how numpy handles A[:,0]

depending on if A is a np.array or np.matrix. In the case of

an array, a 1D array is returned, in the case of a matrix, a 2D Nx1

matrix is returned. Using this matrix seems to confuse matplotlib.

Using np.ravel or np.flatten on the slices fix that problem.

Is there an explanation for this behaviour or should I fill a bug ?

Best regards,

Julien Rebetez

matrix_plot.py (396 Bytes)

LJulien Rebetez :

I've run into a strange behaviour of matplotlib while trying to figure

out why my data was displayed incorrectly.

I'm note quite sure if this is a bug or expected behaviour, but I feel

it's kind of counter-intuitive, so I'm posting here.

...

Now there seem to be a difference on how numpy handles A[:,0]

depending on if A is a np.array or np.matrix. In the case of

an array, a 1D array is returned, in the case of a matrix, a 2D Nx1

matrix is returned. Using this matrix seems to confuse matplotlib.

Using np.ravel or np.flatten on the slices fix that problem.

Is there an explanation for this behaviour or should I fill a bug ?

Don't fill a bug.

Read http://www.scipy.org/Tentative_NumPy_Tutorial , please.

They explain that a slice of a matrix is a matrix, and its "view" is different from what you get for arrays.

But, no need to reshape the stuff. Just use the "array attribute" of the matrix :

pl.scatter(B.A[:,0], B.A[:,1], c='b')

//Here 'A' is the name of the attribute, nothing to do with your array A; by chance it is the same...//

The best

Jerzy Karczmarczuk

Thank you for your answer.

I've read the numpy tutorial and I get it that array and matrices

behave differently.

Now, what I find kind of strange is that the plot I get when directly

feeding the matrices to scatter

doesn't really seem to represent anything.

I think that, if possible, showing an error or a warning would be much

more appropriate than showing

a plot. It would let the user know that the problem is not with her

dataset, but with the plot.

Wouldn't it be possible to simply check the shape in scatter() and

display a warning if it has more than one dimension ?

Best,

Julien

## ···

On Tue, Mar 20, 2012 at 3:52 PM, Jerzy Karczmarczuk <jerzy.karczmarczuk@...3937...> wrote:

LJulien Rebetez :

I've run into a strange behaviour of matplotlib while trying to figure

out why my data was displayed incorrectly.

I'm note quite sure if this is a bug or expected behaviour, but I feel

it's kind of counter-intuitive, so I'm posting here.

...

Now there seem to be a difference on how numpy handles A[:,0]

depending on if A is a np.array or np.matrix. In the case of

an array, a 1D array is returned, in the case of a matrix, a 2D Nx1

matrix is returned. Using this matrix seems to confuse matplotlib.

Using np.ravel or np.flatten on the slices fix that problem.

Is there an explanation for this behaviour or should I fill a bug ?

Don't fill a bug.

Read http://www.scipy.org/Tentative_NumPy_Tutorial , please.

They explain that a slice of a matrix is a matrix, and its "view" is

different from what you get for arrays.

But, no need to reshape the stuff. Just use the "array attribute" of the

matrix :

pl.scatter(B.A[:,0], B.A[:,1], c='b')

//Here 'A' is the name of the attribute, nothing to do with your array

A; by chance it is the same...//

The best

Jerzy Karczmarczuk

------------------------------------------------------------------------------

This SF email is sponsosred by:

Try Windows Azure free for 90 days Click Here

http://p.sf.net/sfu/sfd2d-msazure

_______________________________________________

Matplotlib-users mailing list

Matplotlib-users@lists.sourceforge.net

https://lists.sourceforge.net/lists/listinfo/matplotlib-users

I think this is more of an issue that we simply never tested for matrices. As for the shape tests, scatter can take 2d arrays, iirc, with each column being a different color. (again, all from my memory, which would easily fail memcheck).

Ben Root

## ···

On Tuesday, March 20, 2012, Julien Rebetez <julien.rebetez@…287…> wrote:

Thank you for your answer.

I’ve read the numpy tutorial and I get it that array and matrices

behave differently.

Now, what I find kind of strange is that the plot I get when directly

feeding the matrices to scatter

doesn’t really seem to represent anything.

I think that, if possible, showing an error or a warning would be much

more appropriate than showing

a plot. It would let the user know that the problem is not with her

dataset, but with the plot.

Wouldn’t it be possible to simply check the shape in scatter() and

display a warning if it has more than one dimension ?

Best,

Julien