[Matplotlib-users] Looking for feedback on figures using matplotlib and jupyter notebook

[Andreas Mueller]

Hi all.

This is about a joint jupyter-notebook / matplotlib problem I've been
thinking about.
So I'm writing a book using jupyter-notebook, and all my figures are
generated using matplotlib.

In books, there is usually a figure caption with a running number and
some description.
From what I read, the best way to add captions is just using plt.text.
However, the caption should probably be in the markup,
not in a rendered PNG. I'm not sure if changing the backend might help,
but that probably doesn't make the notebook happy?

The other problem is that I want to have running numbers that I can
refer to by a tag (as you would in latex).
That is more of a notebook problem, though.

Any feedback would be very welcome.

Cheers,
Andy

[Fernando Perez]

Post by Andreas Mueller
Hi all.
This is about a joint jupyter-notebook / matplotlib problem I've been
thinking about.
So I'm writing a book using jupyter-notebook, and all my figures are
generated using matplotlib.
In books, there is usually a figure caption with a running number and
some description.
From what I read, the best way to add captions is just using plt.text.
However, the caption should probably be in the markup,
not in a rendered PNG. I'm not sure if changing the backend might help,
but that probably doesn't make the notebook happy?
The other problem is that I want to have running numbers that I can
refer to by a tag (as you would in latex).
That is more of a notebook problem, though.
Any feedback would be very welcome

I've been wanting to do something about this problem for a while, but
haven't had the cycles to work on it... Here's my current idea, perhaps I
can goad you into implementing it :)

I think that IPython.display should provide a Figure object, capable of
wrapping any input image (with nice code to automatically swallow a
matplotlib figure without asking the user to convert it to an image first),
and taking an optional caption.

Figure() would then produce as output the displayed image but with a bit of
nice CSS to center it on the page, along with the caption.

The trick is to send the entire data bundle correctly structured so that,
at the other end, nbconvert could recognize these figures as such, and not
only produce nice HTML, but more importantly, push them into the LaTeX
output with the correct call to \figure, including \caption as well as size
and placement specifiers.

The signature of Figure() might be something like

def Figure(fig, caption=None, width=None, height=None,
latex_placement=None):


I would try implementing this first as a standalone tool, and once it's
been tested enough in real-world usage with both HTML and LaTeX output from
nbconvert, it could be merged in. I suspect it's going to take a few
iterations to get it right.

But it's not particularly hard, and someone working on a book would be the
perfect candidate to have enough test cases to be able to iterate until
happy ;)

If you think you want to take a stab at this, don't hesitate to ping us on
the jupyter list. We can help with some of the more obscure parts of
getting this to work on nbconvert (and there may be things I've overlooked
in the sketch above).

Cheers,

f

[Benjamin Root]

In mpl, our figure objects get numbers assigned to them by default, but
they can also be strings. These labels are used in the figure window title
bar. Perhaps that existing data could be hijacked? Admittedly, most people
use the string name to give nice short names to their figures, so maybe
those names could be the "tag" name in latex? So, all we would need is some
way to supply the actual caption string.

Ben Root

Post by Fernando Perez

Post by Andreas Mueller
Hi all.
This is about a joint jupyter-notebook / matplotlib problem I've been
thinking about.
So I'm writing a book using jupyter-notebook, and all my figures are
generated using matplotlib.
In books, there is usually a figure caption with a running number and
some description.
From what I read, the best way to add captions is just using plt.text.
However, the caption should probably be in the markup,
not in a rendered PNG. I'm not sure if changing the backend might help,
but that probably doesn't make the notebook happy?
The other problem is that I want to have running numbers that I can
refer to by a tag (as you would in latex).
That is more of a notebook problem, though.
Any feedback would be very welcome

I've been wanting to do something about this problem for a while, but
haven't had the cycles to work on it... Here's my current idea, perhaps I
can goad you into implementing it :)
I think that IPython.display should provide a Figure object, capable of
wrapping any input image (with nice code to automatically swallow a
matplotlib figure without asking the user to convert it to an image first),
and taking an optional caption.
Figure() would then produce as output the displayed image but with a bit
of nice CSS to center it on the page, along with the caption.
The trick is to send the entire data bundle correctly structured so that,
at the other end, nbconvert could recognize these figures as such, and not
only produce nice HTML, but more importantly, push them into the LaTeX
output with the correct call to \figure, including \caption as well as size
and placement specifiers.
The signature of Figure() might be something like
def Figure(fig, caption=None, width=None, height=None,
I would try implementing this first as a standalone tool, and once it's
been tested enough in real-world usage with both HTML and LaTeX output from
nbconvert, it could be merged in. I suspect it's going to take a few
iterations to get it right.
But it's not particularly hard, and someone working on a book would be the
perfect candidate to have enough test cases to be able to iterate until
happy ;)
If you think you want to take a stab at this, don't hesitate to ping us on
the jupyter list. We can help with some of the more obscure parts of
getting this to work on nbconvert (and there may be things I've overlooked
in the sketch above).
Cheers,
f
--
fperez.net-at-gmail: mailing lists only (I ignore this when swamped!)
fernando.perez-at-berkeley: contact me here for any direct mail
------------------------------------------------------------------------------
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
Monitor end-to-end web transactions and take corrective actions now
Troubleshoot faster and improve end-user experience. Signup Now!
http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
_______________________________________________
Matplotlib-users mailing list
https://lists.sourceforge.net/lists/listinfo/matplotlib-users

[Andreas Mueller]

Thanks for your input Fernando.
I thought about cross-posting to Jupyter, but I'm glad you also saw it
here :)
That would help, but not solve all problems.
I guess the Figure could hold a tag for referencing, too. It would be
nice to get a tag and caption from matplotlib.
Maybe Benjamin's reply would help with that. But it sounds like the
figure has a single string attached (which is more the tag).
I guess I can do
IPython.display.Figure(matplotlib_figure, caption="stuff", tag="tag")
That would be acceptable, I think.
But how do I reference that in a markup cell? [maybe I should move that
question to the jupyter list, though]

Post by Andreas Mueller
Hi all.
This is about a joint jupyter-notebook / matplotlib problem I've been
thinking about.
So I'm writing a book using jupyter-notebook, and all my figures are
generated using matplotlib.
In books, there is usually a figure caption with a running number and
some description.
From what I read, the best way to add captions is just using plt.text.
However, the caption should probably be in the markup,
not in a rendered PNG. I'm not sure if changing the backend might help,
but that probably doesn't make the notebook happy?
The other problem is that I want to have running numbers that I can
refer to by a tag (as you would in latex).
That is more of a notebook problem, though.
Any feedback would be very welcome
I've been wanting to do something about this problem for a while, but
haven't had the cycles to work on it... Here's my current idea,
perhaps I can goad you into implementing it :)
I think that IPython.display should provide a Figure object, capable
of wrapping any input image (with nice code to automatically swallow a
matplotlib figure without asking the user to convert it to an image
first), and taking an optional caption.
Figure() would then produce as output the displayed image but with a
bit of nice CSS to center it on the page, along with the caption.
The trick is to send the entire data bundle correctly structured so
that, at the other end, nbconvert could recognize these figures as
such, and not only produce nice HTML, but more importantly, push them
into the LaTeX output with the correct call to \figure, including
\caption as well as size and placement specifiers.
The signature of Figure() might be something like
def Figure(fig, caption=None, width=None, height=None,
I would try implementing this first as a standalone tool, and once
it's been tested enough in real-world usage with both HTML and LaTeX
output from nbconvert, it could be merged in. I suspect it's going to
take a few iterations to get it right.
But it's not particularly hard, and someone working on a book would be
the perfect candidate to have enough test cases to be able to iterate
until happy ;)
If you think you want to take a stab at this, don't hesitate to ping
us on the jupyter list. We can help with some of the more obscure
parts of getting this to work on nbconvert (and there may be things
I've overlooked in the sketch above).
Cheers,
f
--
fperez.net-at-gmail: mailing lists only (I ignore this when swamped!)
fernando.perez-at-berkeley: contact me here for any direct mail

[Fernando Perez]

Post by Andreas Mueller
Thanks for your input Fernando.
I thought about cross-posting to Jupyter, but I'm glad you also saw it
here :)
That would help, but not solve all problems.
I guess the Figure could hold a tag for referencing, too. It would be nice
to get a tag and caption from matplotlib.
Maybe Benjamin's reply would help with that. But it sounds like the figure
has a single string attached (which is more the tag).
I guess I can do
IPython.display.Figure(matplotlib_figure, caption="stuff", tag="tag")
That would be acceptable, I think.

Yes, I'd forgotten about the label ("label" is the LaTeX name for what
you're calling "tag" here).

Post by Andreas Mueller
But how do I reference that in a markup cell? [maybe I should move that
question to the jupyter list, though]

Yup, this is the slightly trickier part. A sketch of the solution, we need
to:

- generate a local anchor element for the labeled output. That's the
easier part, it would be the job of the displayed output from this
hypothetical Figure() object.

It needs to wrap the output in `<a name=label>... </a>`.

- For markdown referencing, you simply do [link text](#label).

- The problem would be latex conversion: by default, the above is converted
to

\protect\hyperlink{label}{link text}

where as you want a \ref{label} call instead.

- You also want this to generate the

Internal cross-referencing is one of Markdown's main weaknesses for complex
more document-oriented workflows that aren't purely HTML oriented.
Markdown is really a thin wrapper around HTML, so it doesn't expose the
rich labeling/referencing semantics of rST or LaTeX.

I didn't say this was a done deal, and there might be some tricky edges to
it :)

Cheers

f