Add a "sketch" path filter #1329

mdboom · 2012-10-05T00:14:12Z

Based on the mailing list thread "XKCD style graphs", this adds a sketchy line path filter to matplotlib.

This PR is not complete, but I thought I'd get something up early so we can hack on it together.

To play with this, set the rcParam['path.sketch'] to a tuple (scale, length, randomness), for example (1.0, 128., 16.).

          * *scale*: The amplitude of the wiggle perpendicular to the
            source line, in pixels.

          * *length*: The length of the wiggle along the line, in
             pixels (default 128.0)

          * *randomness*: The scale factor by which the length is
            shrunken or expanded (default 16.0)

This currently only works with the Agg and PDF backends, though it should be easy enough to add to other vector backends by following the same procedure as the PDF backend. The Mac OS-X backend will require more care, but should be possible as it's been added to path_cleanup.cpp.

Beyond this, it would be nice to add a convenience function xkcd_style, or some such, that thickens the lines, removes the north and east spines, possibly changes the font, etc.

mdboom · 2012-10-05T00:17:20Z

Also, I invented my own wiggliness algorithm here based on a randomly progressing sine wave. I think Juergen Hasch's filtered noise approach looks better -- we should probably change this to do that.

pwuertz · 2012-10-05T15:30:13Z

Haha, looks great :)

mdboom · 2012-10-05T15:34:31Z

That image seems to be a broken link, maybe? No worries -- just curious what you've decided to do with it.

pwuertz · 2012-10-05T15:38:32Z

Oops, my bad. I better put it on github...

WeatherGod · 2012-10-05T16:02:36Z

Let's see Matlab match this! :-P

mdboom · 2012-10-05T16:06:58Z

Very cool.

dmcdougall · 2012-10-05T17:03:32Z

Amazing. Unfortunately, this won't make it in 1.2; it's not a bug fix.

:)

dmcdougall · 2012-10-05T17:04:51Z

This is actually great for doing schematic 'big-picture' plots in a talk where a clean looking plot looks 'too scientific'.

NelleV · 2012-10-05T20:14:28Z

lib/matplotlib/rcsetup.py

@@ -337,7 +337,16 @@ def validate_bbox(s):
            return None
        raise ValueError("bbox should be 'tight' or 'standard'")

-
+def validate_sketch(s):
+    if s == 'None' or s is None:


s isn't a very explicit variable name. I'm not sure what the matplotlib's coding convention are regarding to this, but I think the code would gain in readibility with a longer variable name such as sketch.

In cases like this, however, I think using a short variable name is fine. "def validate_sketch(sketch)" looks worse to my eye.

Dunno... Reading the code as a "not matplotlib expert", it didn't seem obvious what s was. I have for rule not to use one letter variables as they are often confusing for beginners, and I want my code to be easily read and modified by anyone. Explicit is better than implicit.

Your suggestion for a name, "sketch" would not help--it is no more descriptive of what "s" is than "s". "s" is not a sketch, it is an argument of unknown type being checked to see if it takes one of several valid forms.

@NelleV @efiring I understand you both have your opinions. They are also both valid opinions. That being said, it seems as though the best course of action here is to agree to disagree.

WeatherGod · 2012-10-06T00:44:53Z

Probably should make it clear in any documentation that this parameter has to be set prior to any plotting calls, much like with setting font sizes and such.

WeatherGod · 2012-10-06T01:06:01Z

sweet! add the plt.rc('path', sketch=(1.0, 128.0, 16.0)) towards the beginning of the animate_decay.py example for some fun!

http://youtu.be/PRRRvIRWk9M

WeatherGod · 2012-10-06T01:36:19Z

And doing it to the subplots.py example is really neat!

http://youtu.be/jx4D2NGqkd0

And it also works for 3D plots as well:

certik · 2012-10-06T02:36:16Z

I don't have time to test this now, but I just want to say that I like this a lot!

glyg · 2012-10-06T05:32:02Z

That's really nice! Now fetch the Ipython crew so we have a whole xkcd python notebook

WeatherGod · 2012-10-06T14:15:16Z

Ok, some bugs:

The GraphicsContextBase doesn't copy the ._sketch attribute when creating a copy of itself.
Artist.update_from() doesn't copy the ._sketch attribute (as well as some others...) (side note, some of the setters in Artist should probably be calling pchanged()).
Even after these changes, I still can not seem to get various collection objects to sketch, so somewhere in chain, the sketch params are not getting propagated.

takluyver · 2012-10-06T21:17:19Z

@glyg: A notebook for you: view | download

(Not my doing - I just saw a link to it)

jakevdp · 2012-10-06T23:22:40Z

Another notebook with some stabs in this direction (though done at the patch level in matplotlib):
http://nbviewer.ipython.org/url/jakevdp.github.com/downloads/notebooks/XKCD_plots.ipynb
Would it be possible for the sketch filter to automatically add a white background line so that there's the bit of space between line crossings?

WeatherGod · 2012-10-07T01:01:00Z

On Saturday, October 6, 2012, Jacob Vanderplas wrote:

Another notebook with some stabs in this direction (though done at the
patch level in matplotlib):

http://nbviewer.ipython.org/url/jakevdp.github.com/downloads/notebooks/XKCD_plots.ipynb
Would it be possible for the sketch filter to automatically add a white
background line so that there's the bit of space between line crossings?

No, that is going to have to be done as a path effect or as an offset
transform.

Nikratio · 2012-11-18T00:01:55Z

@WeatherGod: In your animation examples, even lines that are not being changed (e.g. x and y axis) are constantly wobbling around. Did you do that deliberately, or is that an unintended side effect? To me it looks very irritating.

WeatherGod · 2012-11-18T00:25:40Z

On Saturday, November 17, 2012, Nikolaus Rath wrote:

@WeatherGod https://github.com/WeatherGod: In your animation examples,
even lines that are not being changed (e.g. x and y axis) are constantly
wobbling around. Did you do that deliberately, or is that an unintended
side effect? To me it looks very irritating.

The filter is applied whenever the coordinates of the path is accessed. It
is a natural consequence of the implementation. I rather like it. It
reminds me of some cartoons I remember. We would have to figure out a way
to make the randomness be applied once (maybe on-demand?).

Tillsten · 2013-03-04T20:40:42Z

Any reason not to merge this?

mdboom · 2013-03-05T18:06:17Z

The api for using this is horrible. It needs to have proper kwarg or rcparam access before its mergeable. I've sort of dropped the ball on this one.

WeatherGod · 2013-03-05T18:17:10Z

Also note the bugs that I have pointed out earlier. I think this was a great and innovative proof-of-concept, but it is probably best to re-work this.

A thought I did have about a possible way to address @Nikratio 's concerns (and would enable proper unit-testing) is an ability to cache a seed number for a path. That way, the random seed would be re-used at each re-draw, resulting in the same sketch every time. This seed could be automatically determined and/or explicitly set.

pelson · 2013-03-06T11:34:17Z

FWIW, if my memory serves me, I have a pretty good implementation of this which uses transforms only (with just a couple of bug fixes and if I remember correctly, a single monkey patch). The path filter stuff sounds very useful in this case, but it also sounds like a lot of work and I wonder how widely used it would really be?

certik · 2013-03-06T11:51:43Z

I just wanted to say that it would be very nice to have this in matplotlib,
I think that a lot of people were waiting for this to get in. But I
understand of course that if the code is hackish, it's not worth the pain.

Ondrej

On Wed, Mar 6, 2013 at 12:34 PM, Phil Elson notifications@github.comwrote:

FWIW I have a pretty good implementation of this which uses transforms
only (with just a couple of bug fixes and if I remember correctly, a single
monkey patch). The path filter stuff sounds very useful in this case, but
it also sounds like a lot of work and I wonder how widely used it would
really be?

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/1329#issuecomment-14494827
.

pelson · 2013-03-06T11:56:17Z

Thanks for the info Ondrej.

I think that a lot of people were waiting for this to get in.

Is it purely for XKCD style plots which show data in a less clinical, semantic form, or are there other usecases which make use of the path filter functionality?

pelson · 2013-05-14T09:06:02Z

lib/matplotlib/backend_bases.py

@@ -1003,6 +1005,42 @@ def get_hatch_path(self, density=6.0):
            return None
        return Path.hatch(self._hatch, density)

+    def get_sketch_params(self):
+        """


Not an action: We should think about removing the duplicated docstings (duplicated in Artist) from here.

pelson · 2013-05-14T09:32:57Z

Mike: This is looking good. Please only add commits on top of c288df2 (rather than rebasing / modifying any history) and I would be happy to merge after some of the actions have been addressed/discussed.

I love the xkcd rendered version of the docs - very nicely done 😄

…e-created text objects

mdboom · 2013-05-14T15:48:31Z

@pelson: Thanks for the feedback. I believe I have addressed all of your comments. The context manager took some work because the font.* rcParams are "dynamic". I've changed this so that they only affect to-be-created text objects and not ones already created (this brings them in line with most other rcParams). That's a pretty significant change, but it surprisingly doesn't break any tests. Still something to take some caution about.

Out of curiosity, why the rebase warning?

pelson · 2013-05-14T16:19:46Z

Out of curiosity, why the rebase warning?

I didn't want to re-read the whole lot again 😄.

This all looks good to me. Very nice work 😄

Add a "sketch" path filter for XKCD style graphics.

pelson · 2013-05-15T10:41:54Z

src/_path.cpp

@@ -58,7 +58,7 @@ class _path_module : public Py::ExtensionModule<_path_module>
        add_varargs_method("convert_path_to_polygons", &_path_module::convert_path_to_polygons,
                           "convert_path_to_polygons(path, trans, width, height)");
        add_varargs_method("cleanup_path", &_path_module::cleanup_path,
-                           "cleanup_path(path, trans, remove_nans, clip, snap, simplify, curves)");
+                           "cleanup_path(path, trans, remove_nans, clip, snap, simplify, curves, sketch_params)");


~~@mdboom - I don't this this string is correct.~~

NelleV reviewed Oct 5, 2012
View reviewed changes

ghost assigned mdboom Apr 15, 2013

mdboom mentioned this pull request Apr 16, 2013

patheffects for Line2d object : rebase of #1015 #1909

Closed

leejjoon added 2 commits May 13, 2013 15:28

implement path_effects for Line2D objects

5c26a7e

modify patheffect_demo.py to demonstrate patheffects for Line2D objects

99e2b5e

pelson reviewed May 14, 2013
View reviewed changes

mdboom added 5 commits May 14, 2013 08:53

Credit basis for plots

edb1526

Change behavior of font.* rcParams so they take effect only on to-b…

cc61700

…e-created text objects

Miscellaneous fixes mentioned in the PR

9652004

Provide xkcd as a context manager

a274a41

Add credit to figures themselves

4a61f50

pelson added a commit that referenced this pull request May 14, 2013

Merge pull request #1329 from mdboom/wiggles

6b442b8

Add a "sketch" path filter for XKCD style graphics.

pelson merged commit 6b442b8 into matplotlib:master May 14, 2013

pelson reviewed May 15, 2013
View reviewed changes

mdboom deleted the wiggles branch August 7, 2014 13:52

tacaswell mentioned this pull request Dec 5, 2016

API: convert string fontsize to points immediately #7007

Merged

tacaswell mentioned this pull request Jul 8, 2017

Prefer to the GraphicsContext public API when possible. #8848

Merged

6 tasks

Add a "sketch" path filter #1329

Add a "sketch" path filter #1329

Conversation

mdboom commented Oct 5, 2012

mdboom commented Oct 5, 2012

pwuertz commented Oct 5, 2012

mdboom commented Oct 5, 2012

pwuertz commented Oct 5, 2012

WeatherGod commented Oct 5, 2012

mdboom commented Oct 5, 2012

dmcdougall commented Oct 5, 2012

dmcdougall commented Oct 5, 2012

NelleV Oct 5, 2012

Choose a reason for hiding this comment

efiring Oct 5, 2012

Choose a reason for hiding this comment

NelleV Oct 9, 2012

Choose a reason for hiding this comment

efiring Oct 9, 2012

Choose a reason for hiding this comment

dmcdougall Oct 9, 2012

Choose a reason for hiding this comment

WeatherGod commented Oct 6, 2012

WeatherGod commented Oct 6, 2012

WeatherGod commented Oct 6, 2012

certik commented Oct 6, 2012

glyg commented Oct 6, 2012

WeatherGod commented Oct 6, 2012

takluyver commented Oct 6, 2012

jakevdp commented Oct 6, 2012

WeatherGod commented Oct 7, 2012

Nikratio commented Nov 18, 2012

WeatherGod commented Nov 18, 2012

Tillsten commented Mar 4, 2013

mdboom commented Mar 5, 2013

WeatherGod commented Mar 5, 2013

pelson commented Mar 6, 2013

certik commented Mar 6, 2013

pelson commented Mar 6, 2013

pelson May 14, 2013

Choose a reason for hiding this comment

pelson commented May 14, 2013

mdboom commented May 14, 2013

pelson commented May 14, 2013

pelson May 15, 2013

Choose a reason for hiding this comment