Fix legend overlapping with lines on causal impact plots

[drbenvincent]

In a number of situations, the legend is overlapping with data in the causal impact plots. Here's an example:

• Create a kwarg in the plot method which allows users to specify that they want the legend outside the plot. It might take some experimentation to get this right. For example, does it look best to put the legend on the side of the plots (but this may look strange) or perhaps under the bottom?

[anevolbap]

Adding something like this should work:
sns.move_legend(ax[0], "upper left", bbox_to_anchor=(1, 1))

Maybe figsize can be adjusted to make wider as before 🤔

[drbenvincent]

Sorry for slow reply @anevolbap - notification got buried.

I think that could be good, though I'd prefer direct matplotlib manipulation rather than seaborn, I think it's still just a 1 liner.

I'm also wondering about placing the legend at the bottom. Maybe we can have a kwarg for the plot method where we can ask for "bottom" or "right" for example.

I can't remember now, but I believe we pass back the figure and axis handles, so if a user wanted fine-grained control then they could do that after the call to result.plot().

[anevolbap]

No worries, @drbenvincent. I'm quite busy over here too. I'll work on your comments. Thanks!

[OriolAbril]

Instead of using seaborn's move_legend (which actually deletes the legend and creates a new one under the hood), you can pass these same arguments to the call to ax.legend. Would the goal be to allow users to pass arbitrary kwargs to ax.legend or would it be preferrable to have a single legend kwarg that takes a few strings representing common placements?

[drbenvincent]

I don't have strong feelings about it, but my slight preference would be towards allowing the user more control. As long as we have an example (and info in the plot docstrings) then it shouldn't be too tricky for people to figure out.