RcParams instances for matplotlib.style.use #3795

chebee7i · 2014-11-14T01:01:11Z

External libraries may want to prepare styles dynamically. For example, brewer2mpl provides various colormaps. While it could create it's own context manager for temporarily changing the color_cycle, I think it makes more sense if matplotlib.style.use can work directly with RcParams instances.

>>> with matplotlib.style.context(matplotlib.RcParams(**{'text.usetex':False})):
...     print matplotlib.rcParams['text.usetex']
...      
False

tacaswell · 2014-11-14T06:09:05Z

There is already a rcparam context manager matplotlib.rc_contex iirc.

chebee7i · 2014-11-14T07:04:21Z

The intended use-cases are a bit different though.

matplotlib.rc_context seems to be intended for loading complete rcParams from file (ala use_default_template=True) or for updating the existing rcParams from a dictionary (and doing both in combination). This will certainly work for the use case I described.

On the other hand, matplotlib.style.context is tailored towards loading and chaining styles (ala use_default_template=False). This is more flexible and with this PR, we can do something like:

rc = matplotlib.RcParams([('text.usetext', True)])
with matplotlib.style.context(['ggplot2', rc, '/path/to/myfile.mplstyle']):
    pass

To get the same functionality with matplotlib.rc_context, you'd have to 1) use nested with rc_context(rc=blah) blocks, 2) merge the dictionaries beforehand, or use 3) contextlib.nested.

tacaswell · 2014-11-14T07:33:05Z

I don't really understand, rc_context works as expected with just a dict as an arg and nests just fine:

import matplotlib
matplotlib.rcParams['font.size'] = 314
print matplotlib.rcParams['text.usetex'], matplotlib.rcParams['font.size']
with matplotlib.rc_context(rc={'text.usetex': True}):
    print matplotlib.rcParams['text.usetex'], matplotlib.rcParams['font.size']
    with matplotlib.rc_context(rc={'text.usetex': False}):
        print matplotlib.rcParams['text.usetex'], matplotlib.rcParams['font.size']
        with matplotlib.rc_context(rc={'font.size': 12}):
            print matplotlib.rcParams['text.usetex'], matplotlib.rcParams['font.size']
        print matplotlib.rcParams['text.usetex'], matplotlib.rcParams['font.size']
    print matplotlib.rcParams['text.usetex'], matplotlib.rcParams['font.size']
print matplotlib.rcParams['text.usetex'], matplotlib.rcParams['font.size']

The use case in your second post does make this change more compelling.

attn @tonysyu

tacaswell · 2014-11-14T07:33:34Z

lib/matplotlib/style/core.py

@@ -45,16 +45,19 @@ def use(name):

    Parameters
    ----------
-    name : str or list of str
+    name : str, list of str, or RcParams


should accept a dict objects

chebee7i · 2014-11-14T08:10:53Z

@tacaswell Yes, that's all I was saying: rc_context requires nesting in order to chain, whereas style.context chains for you (and also works with named styles). I'll go ahead and make those changes.

chebee7i · 2014-11-14T09:01:25Z

I ended up trying:

    if cbook.is_string_like(name) or hasattr(name, 'keys'):
        name = [name]

CPython seems to require a keys attribute in order for update() to work at all. From your suggestion, another alternative could be:

if cbook.is_string_like(name):
    name = [name]
else:
    try:
        mpl.rcParams.update(name)
    except:
        # Correctly catches when name is a list, but bad dicts pass through without an error.
        pass
    else:
        name = []

for style in name:
    ...

Let me know what you think is better, or if you have a better idea.

chebee7i · 2014-11-14T19:32:00Z

Unit test failures seem unrelated now: FAIL: matplotlib.tests.test_axes.test_pcolormesh.test

tacaswell · 2014-11-16T01:47:41Z

lib/matplotlib/style/core.py

        name = [name]

    for style in name:
+        if not cbook.is_string_like(style):


It seems very odd to test not is_string_like in a is_string_like.

The not is_string_like was after (not within) a previous call to is_string_like. The only reason I chose the negated form for the second call was to avoid having to indent everything an extra level. We could also just do the extra indentation:

if cbook.is_string_like(style): if style in library: mpl.rcParams.update(library[style]) else: try: rc = rc_params_from_file(style, use_default_template=False) mpl.rcParams.update(rc) except: msg = ("'%s' not found in the style library and input is " "not a valid URL or path. See `style.available` for " "list of available styles.") raise ValueError(msg % style) else: mpl.rcParams.update(style)

Preferred?

Blah, sorry, I misread the diffs, disregard this comment.

tonysyu · 2014-11-16T20:22:01Z

lib/matplotlib/style/core.py

-        Name of style or path/URL to a style file. For a list of available
-        style names, see `style.available`. If given a list, each style is
-        applied from first to last in the list.
+    name : str, dict, list of str and/or dict


With this addition, the name of this argument is a bit misleading (actually, it was a bit misleading before, too). Maybe something like style_spec, or something similar. (By "spec", I mean specifier.)

Also, the description is a bit difficult to read because of all the different options. Maybe something like:

style_spec : str, dict, or list str: The name of a style or a path/URL to a style file. For a list of available style names, see `style.available`. dict: Dictionary with valid key/value pairs in `matplotlib.rcParams`. list: List of style specifiers (str or dict) applied from first to last in the list.

I'll go for "style" and see what you guys think.

chebee7i · 2014-11-24T22:13:43Z

Anything else needed for this?

tacaswell · 2014-11-24T22:24:07Z

Some what out side of the scope of this PR, but this needs to have the bit of logic added to the rc_context recently to make sure that the rcParams is rolled back to it's previous state if something raises an exception in update.

tacaswell · 2014-11-24T22:24:45Z

lib/matplotlib/style/core.py

-        style names, see `style.available`. If given a list, each style is
-        applied from first to last in the list.
+    style : str, dict, or list
+        str: The name of a style or a path/URL to a style file. For a list of


Can you make this a rst table? I don't think this is going to render very will in the html docs.

Oops, I guess this doesn't render correctly. Personally, I think a simple restructuredtext-list would be easier to read in plain text.

tacaswell · 2014-11-24T22:29:40Z

I would like @tonysyu to sign off on this before it gets merged as well.

chebee7i · 2014-11-25T03:18:41Z

@tacaswell Regarding errors on updating, all the code uses the update method on RcParams. This does not provide validation, so while we might catch some errors, we won't catch the errors we really want to catch. Should the updates loop over key/value pairs instead? [This also affects style.context.]

chebee7i · 2014-11-25T03:21:45Z

I tried a table. In plain text it looks ok. Let me know what you think. Too bad sphinx isn't more flexible here.

tonysyu · 2014-11-25T03:28:59Z

The table looks fine too. 👍

tacaswell · 2014-11-25T03:31:39Z

update does validate now, #3564

See #3752 for the issue where
the error handling was added to the main context manager.

On Mon Nov 24 2014 at 10:21:50 PM chebee7i [email protected] wrote:

I tried a table. In plain text it looks ok. Let me know what you think.

—
Reply to this email directly or view it on GitHub
#3795 (comment)
.

chebee7i · 2014-11-25T04:13:59Z

Rebased and added the logic for restore rcParam state.

ENH : RcParams/dict as input to matplotlib.style.use

tacaswell · 2014-11-25T05:19:22Z

@chebee7i Thanks!

If you are up for it, some more work done to unify the style and rcparam code. It seems a little silly to have two almost identical context managers and the rcparams code in general could use some cleaning up.

I am not sure the best way to go about doing that (to maintain back compatibility and avoid circular dependencies).

tacaswell reviewed Nov 14, 2014
View reviewed changes

tacaswell added this to the v1.5.x milestone Nov 14, 2014

tacaswell added the status: needs revision label Nov 14, 2014

chebee7i force-pushed the style-rcparams branch 3 times, most recently from 4cddc55 to 9cdc0c6 Compare November 14, 2014 08:53

tacaswell reviewed Nov 16, 2014
View reviewed changes

tacaswell added status: needs review and removed status: needs revision labels Nov 16, 2014

tonysyu reviewed Nov 16, 2014
View reviewed changes

tacaswell added status: needs revision and removed status: needs revision status: needs review labels Nov 18, 2014

tacaswell reviewed Nov 24, 2014
View reviewed changes

tacaswell added status: needs review and removed status: needs revision labels Nov 24, 2014

chebee7i added 8 commits November 24, 2014 22:11

Allow style.use() to accept RcParams instances.

4621942

Remove isinstance() calls in mpl.styles.use().

bbeb742

Add unit test for style.context() using a dict.

32a27ff

PEP 8.

998ac7e

Update docstring for style.use() and style.context().

17e8f9c

Add more informative unit test names. Style cleanup.

17bf8d4

Use table for docstring.

636241c

Reset style values on any error.

0fe5f25

chebee7i force-pushed the style-rcparams branch from 429e51d to 0fe5f25 Compare November 25, 2014 04:11

Fixing "line too long (80 > 79 characters)". :-/

ade114f

chebee7i force-pushed the style-rcparams branch from b4e6af0 to ade114f Compare November 25, 2014 04:20

Missing import for SkipTest.

0a64840

tacaswell added a commit that referenced this pull request Nov 25, 2014

Merge pull request #3795 from chebee7i/style-rcparams

980e4cd

ENH : RcParams/dict as input to matplotlib.style.use

tacaswell merged commit 980e4cd into matplotlib:master Nov 25, 2014

tacaswell removed the status: needs review label Nov 25, 2014

Uh oh!

RcParams instances for matplotlib.style.use #3795

RcParams instances for matplotlib.style.use #3795

Uh oh!

Conversation

chebee7i commented Nov 14, 2014

Uh oh!

tacaswell commented Nov 14, 2014

Uh oh!

chebee7i commented Nov 14, 2014

Uh oh!

tacaswell commented Nov 14, 2014

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chebee7i commented Nov 14, 2014

Uh oh!

chebee7i commented Nov 14, 2014

Uh oh!

chebee7i commented Nov 14, 2014

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!

chebee7i commented Nov 24, 2014

Uh oh!

tacaswell commented Nov 24, 2014

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tacaswell commented Nov 24, 2014

Uh oh!

chebee7i commented Nov 25, 2014

Uh oh!

chebee7i commented Nov 25, 2014

Uh oh!

tonysyu commented Nov 25, 2014

Uh oh!

tacaswell commented Nov 25, 2014

Uh oh!

chebee7i commented Nov 25, 2014

Uh oh!

tacaswell commented Nov 25, 2014

Uh oh!

Uh oh!