Added ability to offset errorbars when using errorevery. #6280

jcalbert · 2016-04-08T02:41:54Z

I was having a problem where errorbars for different series would perfectly overlap, as shown in the top plot:

I added the ability to specify the errorevery such that

plt.errorbar(x,y,yerr, errorevery=7) still adds error bars to the 1st, 8th, 15th ... data points but

plt.errorbar(x,y,yerr, errorevery=(7,3)) adds error bars to the 3rd, 10th, 17th ... points.

I apologize if I'm violating some protocol - I haven't contributed to a project like this before.

P.S. The plot above was generated by:

from matplotlib import pyplot as plt
import numpy as np

x1 = np.arange(0,1000.)
x2 = np.arange(0,1000.)

y1 = np.cumsum(np.random.rand(x1.size)-.495)
y2 = np.cumsum(np.random.rand(x2.size)-.505)

ye1 = x1**.5 / 4
ye2 = x2**.5 / 4

dat1 = [x1,y1,ye1]
dat2 = [x2,y2,ye2]


fig = plt.figure(figsize=(12,8))

ax_before = fig.add_subplot(2,1,1); ax_before.set_title('Without offsets')
ax_after  = fig.add_subplot(2,1,2); ax_after.set_title('With offsets')

for x,y,ye in [dat1, dat2]:
    shift = np.all(y==y1) * 20

    ax_before.errorbar(x,y,ye, errorevery=40,capsize=5)
    ax_after.errorbar(x,y,ye, errorevery=(40,shift),capsize=5)

fig.savefig('compare.png')

tacaswell · 2016-04-08T03:08:12Z

lib/matplotlib/axes/_axes.py

+        if not iterable(errorevery):
+            errorevery = (errorevery, 0)
+        try:
+            assert all(int(x)==x for x in errorevery)


I while ago we cleaned all of the asserts out of the main code base and use a if ... raise pattern.

tacaswell · 2016-04-08T03:13:53Z

👍 I like the idea, just left some code-style comments.

Can you also add a test (just a png image comparison should be enough) and a whats_new entry ( https://github.com/matplotlib/matplotlib/tree/master/doc/users/whats_new )

jcalbert · 2016-04-08T17:52:34Z

Done - is there any documentation / examples I should do?

tacaswell · 2016-04-08T17:58:10Z

👍 Looks good to me.

If you want to update the mark-every examples in the gallery that would be good.

QuLogic · 2016-04-08T22:53:24Z

lib/matplotlib/axes/_axes.py

            subsamples the errorbars. e.g., if errorevery=5, errorbars for
-            every 5-th datapoint will be plotted. The data plot itself still
-            shows all data points.
+            every 5-th datapoint will be plotted. If errorevery=(a,b), points


Missing space after comma in tuple.

jcalbert · 2016-04-09T23:20:42Z

In an earlier revision, I had validation of the sort which ensured

errorevery==int(errorevery)
offset==int(offset)
errorevery>=1
offset>=0
but simplified it in accordance with @tacaswell 's suggestion above.

....

Regarding @QuLogic 's note:

Personally, I prefer
(np.arange(N) + offset) % errorevery == 0
over

offset %= errorevery 
np.arange(N) % errorevery == offset

tacaswell · 2016-04-11T01:19:51Z

Sometimes I make bad suggestions 🐑

(np.arange(N) + offset) % errorevery == 0 seems like a nice way to do this.

QuLogic · 2016-04-11T01:24:16Z

That should be - offset.

jcalbert · 2016-04-12T22:31:11Z

Hm, not sure how I broke test_savefig_to_stringio_eps_afm.

jcalbert · 2016-04-12T22:32:37Z

(My bad, I didn't realize "Close and comment" closed the issue)

tacaswell · 2016-04-12T22:33:41Z

That test is known to be flaky, but no one has been able to reproduce it reliably to debug it. We are pretty sure it is an artifact of running nose in parallel + not perfectly isolated tests.

dstansby · 2017-02-06T18:40:11Z

This looks like a cool new feature! @jcalbert could you rebase this on to the current master branch, and then I'll have a look at it in detail?

jcalbert · 2017-02-07T02:12:06Z

@dstansby Alright, it's going to take me a second to learn git rebase though, since I haven't used it before.

tacaswell · 2017-02-07T04:11:22Z

Sorry this got lost in the weeds 😞

tacaswell · 2019-02-24T20:02:58Z

power-cycled to re-run CI.

This looks like a very useful feature (that got a lot of discussion), however from skimming the discussion it looks like there are still outstanding issues with the API being inconsistent with the API of plot?

jcalbert · 2019-04-15T00:17:11Z

IIRC, I think it was getting mysteriously snagged on one of the CI tests but I don't know how to go back and check old results. I think we settled on having errorevery support a subset of the functionality of markevery, but that subset is consistent.

I'll rebase and fix one style issue "(6,3)" -> "(6, 3)" tomorrow

jcalbert · 2019-04-17T16:38:11Z

Should be good now.

tacaswell · 2019-04-20T19:54:45Z

doc/users/whats_new/errorbar_offsets.rst

@@ -0,0 +1,10 @@
+Errorbar plots can shift which points have error bars
+----------------------


The decorator should be the same length as the title.

tacaswell

modulo moving the whats_new file to next_whats_new

tacaswell · 2019-04-20T19:57:52Z

Thanks for sticking with this for so long!

timhoffm · 2019-04-22T14:41:39Z

examples/lines_bars_and_markers/errorbar_subsample.py



 # Now switch to a more OO interface to exercise more features.
-fig, axs = plt.subplots(nrows=1, ncols=2, sharex=True)
+fig, axs = plt.subplots(nrows=1, ncols=3, sharex=True, figsize=(12, 6))


Suggested change

fig, axs = plt.subplots(nrows=1, ncols=3, sharex=True, figsize=(12, 6))

fig, (ax1, ax2) = plt.subplots(nrows=1, ncols=3, sharex=True, figsize=(12, 6))

and using ax1, ax2 below is better than reassigning ax = ax[i] later.

timhoffm · 2019-04-22T14:44:08Z

examples/lines_bars_and_markers/errorbar_subsample.py

@@ -12,22 +12,29 @@

 # example data
 x = np.arange(0.1, 4, 0.1)
-y = np.exp(-x)
+y = np.exp(np.vstack([-1.0 * x, -.5 * x]))


I think, it's easier to follow if you keep the data in two separate variables:

y1 = np.exp(-x) y2 = np.exp(-0.5 * x)

timhoffm · 2019-04-22T14:44:37Z

examples/lines_bars_and_markers/errorbar_subsample.py


 # example variable error bar values
-yerr = 0.1 + 0.1 * np.sqrt(x)
+yerr = 0.1 + 0.1 * np.sqrt(np.vstack([x, x/2]))


Same with y1err, y2err.

timhoffm · 2019-04-22T14:46:52Z

examples/lines_bars_and_markers/errorbar_subsample.py

 ax = axs[0]
-ax.errorbar(x, y, yerr=yerr)
+for i in range(2):
+    ax.errorbar(x, y[i], yerr=yerr[i])


With above changes simpler:

ax1.errorbar(x, y1, yerr=y1err) ax1.errorbar(x, y2, yerr=y2err)

Same for the following for loops.

timhoffm · 2019-04-22T14:53:18Z

lib/matplotlib/axes/_axes.py

+        except TypeError:
+            offset = 0
+
+        int_msg = 'errorevery must be positive integer or tuple of integers'


It's more readable to define the message directly in the context of the ValueError. Also, this allows to make the error message more specific.

if errorevery < 1 or int(errorevery) != errorevery: raise ValueError( 'errorevery must be positive integer or tuple of integers') if int(offset) != offset: raise ValueError('errorevery offset must be an integer >= 0')

jcalbert · 2019-04-23T01:02:14Z

Added @timhoffm 's suggestions, fixed what's new.

timhoffm · 2019-04-26T12:27:39Z

flake8 complains:
./examples/lines_bars_and_markers/errorbar_subsample.py:24:57: W291 trailing whitespace

Doc build is broken due to #14003 (unrelated to this PR). Can be fixed by rebasing.

In addition to adding errorbars on a subset of data points one may add them to (2, 5, 8...) rather than (0, 3, 6...) to avoid overlaps.

made 2-argument errorevery give a more accurate error message

timhoffm · 2019-04-28T16:55:31Z

Thanks a lot for keeping at this! I think this is PR with the longest history I've merged into matplotlib. Anyway, glad that it's in 🎉. Hope to see you back some time.

mdboom added the status: needs review label Apr 8, 2016

tacaswell added this to the 2.1 (next point release) milestone Apr 8, 2016

tacaswell reviewed Apr 8, 2016
View reviewed changes

tacaswell added status: needs revision and removed status: needs review labels Apr 8, 2016

tacaswell added status: needs review and removed status: needs revision labels Apr 8, 2016

QuLogic reviewed Apr 8, 2016
View reviewed changes

jcalbert closed this Apr 12, 2016

mdboom removed the status: needs review label Apr 12, 2016

jcalbert reopened this Apr 12, 2016

mdboom added the status: needs review label Apr 12, 2016

WeatherGod mentioned this pull request Feb 6, 2017

Add errorbars to mplot3d #8031

Merged

dstansby self-assigned this Feb 8, 2017

tacaswell reopened this Feb 24, 2019

tacaswell modified the milestones: v3.1.0, v3.2.0 Feb 24, 2019

jcalbert force-pushed the errorbar-offset branch from 4523509 to 9fd0233 Compare April 16, 2019 00:42

tacaswell added the Release critical For bugs that make the library unusable (segfaults, incorrect plots, etc) and major regressions. label Apr 20, 2019

tacaswell reviewed Apr 20, 2019

View reviewed changes

tacaswell approved these changes Apr 20, 2019

View reviewed changes

timhoffm reviewed Apr 22, 2019

View reviewed changes

timhoffm approved these changes Apr 26, 2019

View reviewed changes

jcalbert added 11 commits April 26, 2019 17:20

Added ability to offset errorbars with errorevery

fee1591

In addition to adding errorbars on a subset of data points one may add them to (2, 5, 8...) rather than (0, 3, 6...) to avoid overlaps.

modified errorbar_subsample example to demonstrate feature

0741bc7

tests to check functionality using check_figures_equal decorator

5394e84

document new features

d75fb49

style compliance fixes

59db2ed

fixed example to match behavior for two-tuple arguments

04d15d6

moved what's new to new what's new

7b16d13

proper length heading line for .rst

f924d4d

unpacked and named y, yerr, and axs (previously iterables) for clarity.

e24476e

moved error message into ValueError

fc0f3b9

made 2-argument errorevery give a more accurate error message

tiny flake8 change

2bfcf8c

jcalbert force-pushed the errorbar-offset branch from 718522c to 2bfcf8c Compare April 26, 2019 21:26

timhoffm merged commit 4507a59 into matplotlib:master Apr 28, 2019

jcalbert deleted the errorbar-offset branch May 3, 2019 23:20

		@@ -0,0 +1,10 @@
		Errorbar plots can shift which points have error bars
		----------------------

	fig, axs = plt.subplots(nrows=1, ncols=3, sharex=True, figsize=(12, 6))
	fig, (ax1, ax2) = plt.subplots(nrows=1, ncols=3, sharex=True, figsize=(12, 6))

Uh oh!

Added ability to offset errorbars when using errorevery. #6280

Added ability to offset errorbars when using errorevery. #6280

Uh oh!

Conversation

jcalbert commented Apr 8, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tacaswell commented Apr 8, 2016

Uh oh!

jcalbert commented Apr 8, 2016

Uh oh!

tacaswell commented Apr 8, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jcalbert commented Apr 9, 2016

Uh oh!

tacaswell commented Apr 11, 2016

Uh oh!

QuLogic commented Apr 11, 2016

Uh oh!

jcalbert commented Apr 12, 2016

Uh oh!

jcalbert commented Apr 12, 2016

Uh oh!

tacaswell commented Apr 12, 2016

Uh oh!

dstansby commented Feb 6, 2017

Uh oh!

jcalbert commented Feb 7, 2017

Uh oh!

tacaswell commented Feb 7, 2017

Uh oh!

tacaswell commented Feb 24, 2019

Uh oh!

jcalbert commented Apr 15, 2019

Uh oh!

jcalbert commented Apr 17, 2019

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tacaswell left a comment

Choose a reason for hiding this comment

Uh oh!

tacaswell commented Apr 20, 2019

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jcalbert commented Apr 23, 2019

Uh oh!

timhoffm commented Apr 26, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

timhoffm commented Apr 28, 2019

Uh oh!

Uh oh!

timhoffm commented Apr 26, 2019 •

edited

Loading