Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Semilog plots with zero values #3998

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
raoulbq opened this issue Jan 13, 2015 · 5 comments
Closed

Semilog plots with zero values #3998

raoulbq opened this issue Jan 13, 2015 · 5 comments
Labels
Release critical For bugs that make the library unusable (segfaults, incorrect plots, etc) and major regressions. status: confirmed bug
Milestone
v2.1

Comments

@raoulbq
Copy link

raoulbq commented Jan 13, 2015

Consider the following test code:

from numpy import *
from matplotlib.pyplot import *


x = arange(-10,10)
y = (1.0 - 1.0/(x**2+1))**20

figure()
semilogy(x, y, "-o")
grid(True)
savefig("test.png")

Correct Plot (the value y=0 at x=0 is masked out from y-log plot):
test_good

Wrong Plot (spurious point at x=0 with y=1e-1):
test_wrong

A git bisect tells me that:

91725d8665975de9a26d631445ab1448b0391e1a is the first bad commit
commit 91725d8665975de9a26d631445ab1448b0391e1a
Author: Joel B. Mohler <[email protected]>
Date:   Thu Aug 21 10:43:28 2014 -0400

    use _fast_from_codes_and_verts in transform code

The commit diff introduces the use of _fast_from_codes_and_verts,
so the issue is probably somewhere deeper inside this function.
@tacaswell tacaswell added this to the v1.4.3 milestone Jan 13, 2015
@tacaswell tacaswell added status: confirmed bug Release critical For bugs that make the library unusable (segfaults, incorrect plots, etc) and major regressions. labels Jan 13, 2015
@tacaswell
Copy link
Member

attn @jbmohler

@tacaswell tacaswell modified the milestones: v1.4.x, v1.4.3, v1.5.x Jan 17, 2015
@tacaswell
Copy link
Member

Retagged this as 1.5.x as this commit is not on the 1.4.x branch phew.

@jbmohler
Copy link
Contributor

Good, I'm also glad to see this is not on the 1.4.x branch. I looked at this a bit, but it's not obvious to me how the optimized constructor is in-equivalent from the normal Path init relevant to this bug. My general comment is that looking at this reminded me how expensive the transform code is in the drawing. In fact, it appears that the transform is run separately/redundantly for the vertices and the path. It seems there are two performance issues -- this transform could be cached for later screen updates (e.g. resizing the window doesn't require recomputing the log transform) and the redundancy alleged in this paragraph.

It may make sense to revert the change and look at the transform code altogether. Certainly that's a daunting prospect.

Perhaps here is as good as any place to make my next mildly outrageous claim. There's a 4-fold (or more) rendering speed improvement for interactive applications (e.g. qtagg backend) to be had by doing the following 3 things:

These things are in order of speed gain so it's difficult for me to get enthusiastic about attacking transform code.

tacaswell added a commit to tacaswell/matplotlib that referenced this issue Jan 17, 2015
@tacaswell
TST : add test for log transform with zero
ef825bf 
example from PR matplotlib#3998
@tacaswell tacaswell mentioned this issue Jan 17, 2015
Closed
@tacaswell
Copy link
Member

@jbmohler I have also been looking at it and am also not seeing what is different. Can you send an email to the devel list re fastaxes?

I am glad some one is looking into ways to make mpl more efficient/better for interactive work!

tacaswell added a commit to tacaswell/matplotlib that referenced this issue Jan 17, 2015
@tacaswell
TST : add test for log transform with zero
9e68161 
example from PR matplotlib#3998
tacaswell added a commit to tacaswell/matplotlib that referenced this issue Jan 17, 2015
@tacaswell
BUG : fix optimized path transforms
24bdc37 
Fixes matplotlib#3998.

The issue is that some transfroms sometimes returned masked arrays from
`transform_non_affine` (ex Log10transform).  The `Path` constructor
fills masked arrays to full arrays with `np.nan`, but the
`_fast_from_codes_and_verts` does not (as it assumes the verts have been
fully validated on the way in).

This adds the maskedarray -> np.nan filled array normalization into
`transform_path_non_affine`.
@tacaswell tacaswell mentioned this issue Jan 29, 2015
Closed
mdboom pushed a commit to mdboom/matplotlib that referenced this issue Jan 29, 2015
@tacaswell @mdboom
TST : add test for log transform with zero
7bc452d 
example from PR matplotlib#3998
@jenshnielsen
Copy link
Member

Fixed by #4050

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
Release critical For bugs that make the library unusable (segfaults, incorrect plots, etc) and major regressions. status: confirmed bug
Projects
None yet
Milestone
v2.1
Development

No branches or pull requests
4 participants
@tacaswell @jenshnielsen @raoulbq @jbmohler