[notebook] Figure legend not shown in inline backend if ouside the box of the axes

[montefra]

In this notebook I use the inline backend and plot two figures with two axes and a figure legend each.
In both cases I adjust the subplot location setting it's right border to 0.7 and put the legend in location right (or 7). In the second case I push the bbox of the legend partially out of the subplot location. Then I save both figures.
The saved figures are correct

But in the notebook it looks like that changing the subplot boundaries with fig.subplots_adjust effectively resize the figure. So in the first case the legend is pushed between the axes, and in the second case the legend is cut.

[Carreau]

Json file seem to be corrupted can't load it (hence 404 nbviewer.. yes not obvious I know)

[montefra]

@Carreau : I updated the issue. I remembered that I have a web address...

[minrk]

It's a bit of a side issue here, but your notebook gist is not valid JSON, how did you upload it?

[montefra]

@minrk : now the notebook should work (converted to html and put on my website)
Before I opened the ipynb with kate, selected all, copied and pasted into gist (this it did work fine yesterday with an other notebook). I've like the feeling that kate did not like the representation of the figures in the ipynb.
They should put an upload button ...

[Carreau]

you can drag and drop on most browser, it allows to upload.

[montefra]

ehm. right it worked, (probably using mostly a the terminal and browser, drag and drop does not come so easy in my mind).

For those of you who prefer nbviewer here is the notebook

[minrk]

The issue is probably the Notebook's use of bbox_inches='tight'.

For instance, try back to back:

fig.savefig("./frominline_figure2.jpg")
fig.savefig("./frominline_figure2tight.jpg", bbox_inches='tight')

and compare the difference. Also what is your matplotlib version?

[montefra]

The figure saved with
fig.savefig("./frominline_figure2tight.jpg", bbox_inches='tight')
appears as the one inlined.

I'm using matplotlib v 1.2.0. I recall seeing somewhere on matplotlib git some PR or Issue related with the inclusion of all artists when doing the tight.

The question then is: why the inline backend enforce bbox_inches='tight'?

[jenshnielsen]

I believe this is identical to matplotlib/matplotlib#1586 which is fixed on the matplotlib 1.2.x and master branches and the fix should make it into the next MPL release. The inline backend uses
bbox_inches='tight' to reduce the amount of white space around the figure. IMHO it would be better to use
plt.tight_layout() that optimises the layout of of the figure to minimize white space but keeps the same figure size instead of clipping. Unfortunately that still has some limitations that makes this impossible at the moment.

[montefra]

From my experience, plt.tight_layout() works only with suplots, not axes.

[montefra]

here is the PR I ment

[jenshnielsen]

I don't think problem is that the bounding box of the legend is not included in the bbox_inches calculation as this PR resolves. The issue that I linked to and which is fixed is that the bbox of the legend is not calculated when the box around the legend is hidden. This is fixed in the mpl 1.2.x and master branch already. Do you have the option to test mpl from the 1.2.x branch?

[montefra]

@jenshnielsen: I see. I can test the mpl master or 1.2.x branch. I'll do it before tonight and let you know

[minrk]

This has been fixed in matplotlib.