Make the legends interactive in bokeh #1024

percygautam · 2020-01-23T18:20:13Z

Fixes part of issue #953 . Make the legends interactive in bokeh.

percygautam · 2020-01-23T18:25:35Z

@ahartikainen @OriolAbril I have started working on the issue and have added the interactive legend feature to densityplot.py. Is this the right way? I'll add legends to other plots in the meantime.

ahartikainen · 2020-01-23T18:45:16Z

It is basically this way. The problem currently is that we probably don't want to have legends on top of the figure, so we would really need a clever way to position the legend outside the plot. Also for multiple figures should share one legend.

Here should be one working example

https://gist.github.com/ahartikainen/506c23b58db06e162da9c188ceab146c

E.g. location probably can take also tuple, but I'm not sure.

percygautam · 2020-01-23T18:57:30Z

@ahartikainen I looked about it and found this open issue Single Legend for GlyphRenderers on Multiple Plots in bokeh. There's a workaround given in the end which we can look for.
There's also a method specified here, which could work for placing legend outside the plot. But the location of where should we put it needs to be discussed.

percygautam · 2020-01-27T21:28:27Z

@ahartikainen , I have tried to adjust the position of legends and this is what I've come up with.

Also, I've tried to optimize the code for better plotting. We just have to fix the position of title, and the legend position looks good to me.

percygautam · 2020-01-27T21:35:42Z

About the other task of multiple figures sharing one legend, I've tried some methods (like passing glyph renders of one plot to another) but it didn't work. Also, in official bokeh, there is an open issue about the same. I'm thinking of trying to use the grid properties for this. Can you please provide any reference for this?
Also, as described in issue #954 , I think we can use subgroups(using dropdown) to select a school and render only one graph at a time. This will also solve the above problem.

ahartikainen · 2020-01-27T21:59:54Z

Looks great.

I think that one-legend-thing might be hard to do. Yeah I think dropdown should be quite "easy" (yeah, they never are) to implement.

Maybe something like dropdown + multiple selections (right)

ahartikainen

LGTM

ahartikainen · 2020-01-27T22:01:33Z

arviz/plots/backends/bokeh/densityplot.py

+                    legend_items[axis_map[label]].append((data_label, plotted))
+
+    for ax1, lg in legend_items.items():
+        legend = Legend(items=lg, location="center_right", orientation="horizontal",)


I think this is a good compromise.

I don't think there is any "universal" way to position legend so it looks good

percygautam · 2020-01-27T22:07:32Z

@ahartikainen So, I'll start working on other plots' legends. Also, I noticed the interactive legends od EnergyPlot are not working. I'll start working on this plot along with other plots having legend( I guess one or two more).

percygautam · 2020-01-27T22:16:10Z

Looks great.

I think that one-legend-thing might be hard to do. Yeah I think dropdown should be quite "easy" (yeah, they never are) to implement.

Maybe something like dropdown + multiple selections (right)

I am thinking of working on dropdowns under the subgroup issue in another PR, as we have to implement those subsections to lot of plots, unlike this legend issue.
The dropdown with single selection will be doable but for multiple selections, I'll try to do but I'm not sure.

ahartikainen · 2020-01-27T22:17:31Z

Sounds good.

No worries if something doesn't work.

percygautam · 2020-01-28T15:20:04Z

@ahartikainen , I have added interactive legends to all other plots (having legends). The discrepancy of energyplot has also been solved and its working fine now.
I have changed the return value of plot_kde, in order to correct energyplot. This change is not affecting any other plot which uses plot_kde, so I don't think it'll cause a problem. We can always add legend the same way in other plots (when required in future) as we did in energyplot.
There is one problem of the title position, which I can't seem to find solution for. We need some way to write the title in same line as legend or is it fine this way?

percygautam · 2020-02-01T06:15:16Z

@ahartikainen I have changed some font sizes. Interactive legends are working fine. This is how they are looking now:

OriolAbril · 2020-02-05T23:37:18Z

arviz/plots/backends/bokeh/kdeplot.py

+    if legend and label is not None:
+        return glyphs


This is what is making the documentation build fail. Maybe bokeh plots can return ax, (extra, elements) or something like this? Maybe a return_glyphs argument only for internal use? @ahartikainen

Yeah, maybe add some argument to bokeh function to return this for internal use. So it would default to false and can not be seen in the main function.

btw did you mean

if legend and label is not None

or

if (legend is not None) and (label is not) None

I have done as @OriolAbril suggested. I just defined return_glyphs as None and returned it along the axes. If the user doesn't specify the legend, it will plot the graph without legend.

OriolAbril

We have to be careful when returning different values from plotting, the examples involving kde will have to be updated too, a priori they will seem to work but thumb generation will fail.

arviz/plots/backends/bokeh/densityplot.py

arviz/plots/backends/bokeh/kdeplot.py

percygautam · 2020-02-10T21:14:53Z

Still getting sphinx error for bokeh_plot_kde. Any workaround?

OriolAbril · 2020-02-10T21:26:55Z

The examples assume the return value is ax, now it is not anymore. You have to update them to ax, _. Docs should also be updated in plot/kdeplot.py

percygautam · 2020-02-10T21:30:21Z

Okay, thanks. I'll do the changes.

We have to be careful when returning different values from plotting, the examples involving kde will have to be updated too, a priori they will seem to work but thumb generation will fail.

Sorry, for not carefully noticing your comment before.

ahartikainen · 2020-02-10T22:35:39Z

I still think our plotting code should return array of axes for the end user and glyphs should be for internal use.

This should be the case until we figure out something more advanced.

percygautam · 2020-02-11T21:03:17Z

I have provided an internal argument that enables/disables the return value of plot_kde. Is this what you're looking for? Also what should be added to its docstring?

ahartikainen

Yes, this looks good.

Maybe add docstring that this is internal use (is there a way for user to return the glyphs?) or that it is for advanced use

OriolAbril

I am not sure about return_glyphs being only in bokeh backend undocumented or allowing advanced users to use it too and therefore making it an arg of general plot_kde and adding it to the docstring. We should probably wait for @ahartikainen to weigh in.

I have commented on how I would handle the docstring if this was eventually chosen, otherwise ignore the comments.

OriolAbril · 2020-02-11T21:10:16Z

arviz/plots/backends/bokeh/kdeplot.py

            ax.x_range.range_padding = ax.y_range.range_padding = 0

    if backend_show(show):
        bkp.show(ax, toolbar_location="above")
+
+    if legend and label is not None and return_glyph:


the if should be only on return_glyph

Then we are getting the problem of unused arguments label and legend. ~~Should I pop them out from kwargs in bokeh backend?~~

OriolAbril · 2020-02-11T21:39:49Z

arviz/plots/kdeplot.py


    Returns
    -------
-    axes : matplotlib axes or bokeh figures
+    axes : matplotlib axes or bokeh figures and glyphs


(docstring comment)

Returns ------- axes : matplotlib.Axes or bokeh.plotting.Figure Object containing the kde plot glyphs : list, optional Bokeh glyphs present in plot. Only provided if ``return_glyph`` is True.

Followed numpy docs like np.unique

OriolAbril · 2020-02-12T15:30:44Z

arviz/plots/backends/bokeh/densityplot.py

+                    legend_items[axis_map[label]].append((data_label, plotted))
+                else:
+                    legend_items[axis_map[label]] = []
+                    legend_items[axis_map[label]].append((data_label, plotted))


Could legend items be a defaultdict?

https://docs.python.org/3/library/collections.html#collections.defaultdict

Yes, it will make the code neat. Thanks, for the suggestion.

OriolAbril · 2020-02-12T15:31:41Z

CHANGELOG.md

@@ -12,6 +12,7 @@
 * Add `skipna` argument to `hpd` and `summary` (#1035)
 * Added `transform` argument to `plot_trace`, `plot_forest`, `plot_pair`, `plot_posterior`, `plot_rank`, `plot_parallel`,  `plot_violin`,`plot_density`, `plot_joint` (#1036)
 * Add `marker` functionality to `bokeh_plot_elpd` (#1040)
+* Add the functionality for `interactive legends` for bokeh plots of `densityplot`, `energyplot` and `essplot` (#1024)


Maybe link to bokeh docs on interactive pegends instead of formatting it like code?

OriolAbril · 2020-02-12T22:32:56Z

LGTM! And I love this feature, I'll probably merge tomorrow. It would be great to include this in next release.

Note: I have edited the description so it won't close the issue. Reason is I think it would be great to test these interactive legends and afterwards try to get them working in plot_trace (definitely in another PR).

percygautam · 2020-02-13T11:22:04Z

Exactly, my thoughts. I was going to ask you about the legend implementation in plot_trace. These interactive legends will make the visualization clean. (And I have added the description earlier such that issue won't close. I thought I'd leave it up for future, if new plots were added.)

ahartikainen

LGTM

Made the legends interactive

56ee7c7

percygautam requested a review from ahartikainen January 23, 2020 18:20

Changed the position of legend to top

046211b

ahartikainen approved these changes Jan 27, 2020

View reviewed changes

Added interactive legends to plots energy, ess, kde

3d9ef3a

linting changes

eb6acd0

percygautam requested a review from ahartikainen January 28, 2020 17:55

percygautam added 2 commits January 28, 2020 23:42

final changes

b1a8904

Minor changes

8cd1301

Merge branch 'master' into bokeh_legend

d4bdbfa

OriolAbril reviewed Feb 5, 2020

View reviewed changes

percygautam added 3 commits February 10, 2020 22:32

Merge branch 'master' into bokeh_legend

0c67b2f

sphinx error correction

d12d9a1

minor changes

7d0b3dd

OriolAbril reviewed Feb 10, 2020

View reviewed changes

arviz/plots/backends/bokeh/densityplot.py Outdated Show resolved Hide resolved

arviz/plots/backends/bokeh/densityplot.py Show resolved Hide resolved

arviz/plots/backends/bokeh/kdeplot.py Outdated Show resolved Hide resolved

Changes

cc9a950

corrected the sphinx error

8f60567

fixed the return axes/glyph issue

29e4f85

ahartikainen approved these changes Feb 11, 2020

View reviewed changes

OriolAbril reviewed Feb 11, 2020

View reviewed changes

percygautam added 5 commits February 12, 2020 15:23

fixed minor errors

806c2a1

Merge branch 'master' into bokeh_legend

eef6d5e

added the change to changelog

7417a5e

corrected the change to changelog

e07ef31

removed the kwargs 'legend' and 'label' from bokeh backend

50ffbd4

OriolAbril approved these changes Feb 12, 2020

View reviewed changes

percygautam added 2 commits February 13, 2020 00:33

optimized the code using defaultdict and changed the changelog

8c0df01

corrected the linting errors

6ff20a8

ahartikainen approved these changes Feb 13, 2020

View reviewed changes

ahartikainen changed the title ~~[WIP] Make the legends interactive in bokeh~~ Make the legends interactive in bokeh Feb 13, 2020

ahartikainen merged commit 0001654 into arviz-devs:master Feb 13, 2020

percygautam deleted the bokeh_legend branch February 17, 2020 20:59

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Make the legends interactive in bokeh #1024

Make the legends interactive in bokeh #1024

percygautam commented Jan 23, 2020 •

edited by OriolAbril

Loading

percygautam commented Jan 23, 2020

ahartikainen commented Jan 23, 2020

percygautam commented Jan 23, 2020

percygautam commented Jan 27, 2020

percygautam commented Jan 27, 2020

ahartikainen commented Jan 27, 2020

ahartikainen left a comment

ahartikainen Jan 27, 2020

percygautam commented Jan 27, 2020

percygautam commented Jan 27, 2020

ahartikainen commented Jan 27, 2020

percygautam commented Jan 28, 2020

percygautam commented Feb 1, 2020

OriolAbril Feb 5, 2020

ahartikainen Feb 6, 2020

percygautam Feb 10, 2020

OriolAbril left a comment

percygautam commented Feb 10, 2020

OriolAbril commented Feb 10, 2020

percygautam commented Feb 10, 2020 •

edited

Loading

ahartikainen commented Feb 10, 2020

percygautam commented Feb 11, 2020

ahartikainen left a comment

OriolAbril left a comment •

edited

Loading

OriolAbril Feb 11, 2020

percygautam Feb 12, 2020 •

edited

Loading

OriolAbril Feb 11, 2020

OriolAbril Feb 12, 2020

OriolAbril Feb 12, 2020

percygautam Feb 12, 2020

OriolAbril Feb 12, 2020

OriolAbril commented Feb 12, 2020

percygautam commented Feb 13, 2020

ahartikainen left a comment

Make the legends interactive in bokeh #1024

Make the legends interactive in bokeh #1024

Conversation

percygautam commented Jan 23, 2020 • edited by OriolAbril Loading

percygautam commented Jan 23, 2020

ahartikainen commented Jan 23, 2020

percygautam commented Jan 23, 2020

percygautam commented Jan 27, 2020

percygautam commented Jan 27, 2020

ahartikainen commented Jan 27, 2020

ahartikainen left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

percygautam commented Jan 27, 2020

percygautam commented Jan 27, 2020

ahartikainen commented Jan 27, 2020

percygautam commented Jan 28, 2020

percygautam commented Feb 1, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

OriolAbril left a comment

Choose a reason for hiding this comment

percygautam commented Feb 10, 2020

OriolAbril commented Feb 10, 2020

percygautam commented Feb 10, 2020 • edited Loading

ahartikainen commented Feb 10, 2020

percygautam commented Feb 11, 2020

ahartikainen left a comment

Choose a reason for hiding this comment

OriolAbril left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

percygautam Feb 12, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

OriolAbril commented Feb 12, 2020

percygautam commented Feb 13, 2020

ahartikainen left a comment

Choose a reason for hiding this comment

percygautam commented Jan 23, 2020 •

edited by OriolAbril

Loading

percygautam commented Feb 10, 2020 •

edited

Loading

OriolAbril left a comment •

edited

Loading

percygautam Feb 12, 2020 •

edited

Loading