figure-introspection.md

jupyter

jupytext kernelspec language_info plotly notebook_metadata_filter text_representation all extension format_name format_version jupytext_version .md markdown 1.2 1.4.2 display_name language name Python 3 python python3 codemirror_mode file_extension mimetype name nbconvert_exporter pygments_lexer version name version ipython 3 .py text/x-python python python ipython3 3.7.7 description display_as language layout name order page_type permalink thumbnail How to dig into and learn more about the figure data structure. file_settings python base Introspecting Figures 36 u-guide python/figure-introspection/ thumbnail/violin.jpg

The Figure Lifecycle

As explained in the Figure Data Structure documentation, when building a figure object with Plotly.py, it is not necessary to populate every possible attribute. At render-time, figure objects (whether generated via Plotly Express or Graph Objects) are passed from Plotly.py to Plotly.js, which is the Javascript library responsible for turning JSON descriptions of figures into graphical representations.

As part of this rendering process, Plotly.js will determine, based on the attributes that have been set, which other attributes require values in order to draw the figure. Plotly.js will then apply either static or dynamic defaults to all of the remaining required attributes and render the figure. A good example of a static default would be the text font size: if unspecified, the default value is always the same. A good example of a dynamic default would be the range of an axis: if unspecified, the default will be computed based on the range of the data in traces associated with that axis.

Introspecting Plotly Express Figures

Figure objects created by Plotly Express have a number of attributes automatically set, and these can be introspected using the Python print() function, or in JupyterLab, the special fig.show("json") renderer, which gives an interactive drilldown interface with search:

import plotly.express as px

fig = px.scatter(x=[10, 20], y=[20, 10], height=400, width=400)
fig.show()
print(fig)

We can learn more about the attributes Plotly Express has set for us with the Python help() function:

help(fig.data[0].__class__.mode)

Accessing Javascript-Computed Defaults

new in 4.10

The .full_figure_for_development() method provides Python-level access to the default values computed by Plotly.js. This method requires the Kaleido package, which is easy to install and also used for static image export.

By way of example, here is an extremely simple figure created with Graph Objects (although it could have been made with Plotly Express as well just like above) where we have disabled the default template for maximum readability. Note how in this figure the text labels on the markers are clipped, and sit on top of the markers.

import plotly.graph_objects as go

fig = go.Figure(
    data=[go.Scatter(
        mode="markers+text",
        x=[10,20],
        y=[20, 10],
        text=["Point A", "Point B"]
    )],
    layout=dict(height=400, width=400, template="none")
)
fig.show()

Let's print this figure to see the very small JSON object that is passed to Plotly.js as input:

print(fig)

Now let's look at the "full" figure after Plotly.js has computed the default values for every necessary attribute.

Heads-up: the full figure is quite long and intimidating, and this page is meant to help demystify things so please read on!

Please also note that the .full_figure_for_development() function is really meant for interactive learning and debugging, rather than production use, hence its name and the warning it produces by default, which you can see below, and which can be suppressed with warn=False.

full_fig = fig.full_figure_for_development()
print(full_fig)

As you can see, Plotly.js does a lot of work filling things in for us! Let's look at the examples described at the top of the page of static and dynamic defaults. If we look just at layout.font and layout.xaxis.range we can see that the static default font size is 12 and that the dynamic default range is computed to be a bit beyond the data range which was 10-20:

print("full_fig.layout.font.size: ", full_fig.layout.font.size)
print("full_fig.layout.xaxis.range: ", full_fig.layout.xaxis.range)

Learning About Attributes

What else can we use this full_fig for? Let's start by looking at the first entry of the data

print(full_fig.data[0])

We see that this is an instance of go.Scatter (as expected, given the input) and that it has an attribute we've maybe never heard of called cliponaxis which by default seems to be set to True in this case. Let's find out more about this attribute using the built-in Python help() function

help(go.Scatter.cliponaxis)

Aha! This explains why in our original figure above, the text was cut off by the edge of the plotting area! Let's try forcing that to False, and let's also use the attribute textposition which we see in the full figure is by default set to "middle center" to get our text off of our markers:

fig.update_traces(cliponaxis=False, textposition="top right")
fig.show()

We can use this technique (of making a figure, and querying Plotly.js for the "full" version of that figure, and then exploring the attributes that are automatically set for us) to learn more about the range of possibilities that the figure schema makes available. We can drill down into layout attributes also:

help(go.layout.XAxis.autorange)

More about Layout

In the figure we introspected above, we had added a scatter trace, and Plotly.js automatically filled in for us the xaxis and yaxis values of that trace object to be x and y, and then also filled out the corresponding layout.xaxis and layout.yaxis objects for us, complete with their extensive set of defaults for gridlines, tick labels and so on.

If we create a figure with a scattergeo trace instead, however, Plotly.js will fill in a totally different set of objects in layout, corresponding to a geo subplot, with all of its defaults for whether or not to show rivers, lakes, country borders, coastlines etc.

import plotly.graph_objects as go

fig = go.Figure(
    data=[go.Scattergeo(
        mode="markers+text",
        lat=[10, 20],
        lon=[20, 10],
        text=["Point A", "Point B"]
    )],
    layout=dict(height=400, width=400,
                margin=dict(l=0,r=0,b=0,t=0),
                template="none")
)
fig.show()
full_fig = fig.full_figure_for_development()
print(full_fig)

If I then set showrivers=True and re-query the full figure, I see that new keys have appeared in the layout.geo object for rivercolor and riverwidth, showing the dynamic nature of these defaults.

fig.update_geos(showrivers=True)
full_fig = fig.full_figure_for_development()
print(full_fig.layout.geo)

Reference

You can learn more about all the available attributes in the plotly figure schema (and read about its high-level structure) or about all the classes and functions in the plotly module.