plotly / dash-sunburst

Dash / React + D3 tutorial: Sunburst diagrams

Please note: this package is intended as a tutorial. It is not recommended for actual use in your apps - for that purpose we have the sunburst trace type in plotly.js / dcc.Graph https://plotly.com/python/sunburst-charts/

Dash Sunburst

This repository demonstrates the principles of combining D3 with React, using a Sunburst chart as an example, and was created from the dash-component-boilerplate template. The Demo uses this Sunburst component to show the contents of a house, with items being added, removed, and resized over time, and letting you zoom in and out of the rooms and items both from within the component itself and from another control.

This component was created primarily as a D3.JS + Dash tutorial. You can use this component in your projects but we are not maintaining it. In fact, we built a first-class Sunburst chart as part of plotly.js and we recommend using this sunburst chart instead: https://plot.ly/python/sunburst-charts/

To run the Dash demo:

1. Clone this repo
2. Run the demo app

python usage.py

3. Open your web browser to http://localhost:8050

Code walkthrough - JavaScript side

Following the structure laid out in the D3 + React tutorial we make two files: d3/sunburst.js for the D3 component and components/Sunburst.react.js for its React/Dash wrapper. Following the dash-component-boilerplate example, this component is then exported using index.js which is imported by the main component in App.js.

Sunburst.react.js

This wrapper simply connects the React component API to the similar structures we create in the D3 component. Excerpting from this file out of order, we see:

Sunburst.propTypes = {
    /**
     * id and setProps are standard for Dash components
     */
    id: PropTypes.string,
    setProps: PropTypes.func,

    /**
     * All the rest are the state of the figure. See the full source for details
     */
    width: PropTypes.number,
    height: PropTypes.number,
    padding: PropTypes.number,
    innerRadius: PropTypes.number,
    transitionDuration: PropTypes.number,
    data: PropTypes.object.isRequired,
    dataVersion: PropTypes.oneOfType([PropTypes.string, PropTypes.number]),
    selectedPath: PropTypes.arrayOf(PropTypes.string),
    interactive: PropTypes.bool
};

In addition to the standard id and setProps props, we insert all the state needed by the D3 component as props of the React wrapper. This gives us type validation - for the most part anyway; this example doesn't validate the structure of data, nor put limits on the numeric fields, but a more production-ready version may want to do this. Note in particular the dataVersion prop. We will use this to avoid having to copy - and diff - the entire data object, which may be large and tedious. Also selectedPath, which is connected to the state of the user interaction with the sunburst, as different parts of the subtree are selected. interactive lets you disable click-to-select nodes, when you want that managed elsewhere.

render() {
    return <div id={this.props.id} ref={el => {this.el = el}} />;
}

In render we just create an empty <div> and store a reference to it in this.el.

componentDidMount() {
    this.sunburst = new SunburstD3(this.el, this.props, figure => {
        const {setProps} = this.props;
        const {selectedPath} = figure;

        if (setProps) { setProps({selectedPath}); }
        else { this.setState({selectedPath}); }
    });
}

componentDidMount instantiates our D3 component, giving it the element to render into, the props for initial render (it will ignore the Dash-specific ones), and a callback to respond to changes from inside that component. A more complex component might emit a variety of events depending on different user interactions, but in the end all that really matters is the current state of the component, not what specifically changed with this event. Here we know the only thing that changed is selectedPath but it would be just as well to call setProps(figure) no matter what event was emitted.

componentDidUpdate() {
    this.sunburst.update(this.props);
}

Whenever the React component gets new props, it simply forwards them on to the D3 component.

App.js

We don't need to know anything about the D3 component in order to use the Sunburst React component in our app - just what's encapsulated in the Sunburst component itself:

constructor() {
    super();
    this.state = {
        transitionDuration: 1000,
        selectedPath: ['living room'],
        dataVersion: 1,
        data: {
            ...
        }
    }
    this.setProps = this.setProps.bind(this);
    this.mutateData = this.mutateData.bind(this);

    this.period = 3;
    this.updateInterval = setInterval(this.mutateData, 1000 * this.period);
}

In the App constructor we start with a seed state for the Sunburst - because this is a simple app where everything is related to that Sunburst, its state is stored in the top level of App.state, but a more complex app would nest it. We also initialize the interval that will periodically edit the data. We don't need to be concerned with the mutateData method, except to know that all it does is call this.setState({data: updatedData})

render() {
    const {data, selectedPath} = this.state;
    const selectedPathStr = selectedPath.join(',');
    const paths = getPathStrs(data, '');
    const options = paths.map(path => (
        <option value={path} key={path}>
            {path.split(',').join('->') || 'root'}
        </option>
    ));
    const selectChange = e => {
        this.setState({selectedPath: e.target.value.split(',')})
    };

    return (
        <div>
            <h2>Sunburst Demo</h2>
            <p>Click a node, or select it in the dropdown, to select a subtree.</p>
            <p>Every {this.period} seconds a node will be added, removed, resized, or renamed</p>
            <Sunburst
                setProps={this.setProps}
                {...this.state}
            />
            <select value={selectedPathStr} onChange={selectChange}>
                {options}
            </select>
        </div>
    )
}

App renders some introductory notes, our Sunburst component, and a dropdown menu that pulls the complete list of paths out of the same state that's used by the Sunburst. You'll notice that whether we select an item by clicking on it directly or via this dropdown, both the dropdown and the Sunburst update.

sunburst.js

Finally, here's the D3 code, all contained in a class we export as SunburstD3.

constructor(el, figure, onChange) {
    const self = this;
    self.update = self.update.bind(self);
    self._update = self._update.bind(self);

    self.svg = d3.select(el).append('svg');
    self.pathGroup = self.svg.append('g');
    self.textGroup = self.svg.append('g')
        .style('pointer-events', 'none');

    self.angularScale = d3.scale.linear().range([0, Tau]);
    self.radialScale = d3.scale.sqrt();
    self.colorScale = d3.scale.category20();
    self.partition = d3.layout.partition()
        .value(d => !d.children && d.size)
        .sort((a, b) => a.i - b.i);

    self.arc = d3.svg.arc()
        .startAngle(d => constrain(self.angularScale(d.x), 0, Tau))
        .endAngle(d => constrain(self.angularScale(d.x + d.dx), 0, Tau))
        .innerRadius(d => Math.max(0, self.radialScale(d.y)))
        .outerRadius(d => Math.max(0, self.radialScale(d.y + d.dy)));

    self.figure = {};

    self.onChange = onChange;

    self.initialized = false;

    self._promise = Promise.resolve();

    self.update(figure);
}

Our constructor does 3 things:

1. Creates the container elements that we'll need no matter what specific diagram we render inside: self.svg is the <svg> element, self.pathGroup will contain the sunburst arcs, and self.textGroup will hold text, added as a separate group so the text will always be in front of the arcs.
2. Pre-calculates d3 helpers that won't change later (self.angularScale through self.arc)
3. Sends the initial figure to self.update. There's also a bit of complication around updating potentially during animations. self._promise is a chain that's added on to whenever a new animation is scheduled, and self.update is an async wrapper around the synchronous self._update, ensuring a new figure is applied only after that chain is complete.

self._update is the meat, so we'll tackle it in pieces:

Figure setup

const oldFigure = self.figure;

// fill defaults in the new figure
const width = figure.width || dflts.width;
const height = figure.height || dflts.height;
// interactive: undefined defaults to true
const interactive = figure.interactive !== false;
const padding = figure.padding || dflts.padding;
const innerRadius = figure.innerRadius || dflts.innerRadius;
const transitionDuration = figure.transitionDuration || dflts.transitionDuration;
const {data, dataVersion} = figure;
const selectedPath = figure.selectedPath || [];

const newFigure = self.figure = {
    width,
    height,
    interactive,
    padding,
    innerRadius,
    transitionDuration,
    data,
    dataVersion,
    selectedPath
};

Here we stash the previous figure as oldFigure and create a new one, inserting default values where values were not provided.

Next comes functions containing our standard D3 code (which was inspired by https://bl.ocks.org/mbostock/4348373 but has been heavily modified, as you can see. Notice that I'm using D3V3 here so some things will change if you're using V4 or V5), but we've broken up the activity by purpose, transitionToNode, updatePaths, and setSize. We'll use these depending on the observed changes. The only items I want to call out within this block are

1. transitionToNode is used in the click callback for our nodes (wrapped up with animation management code).
2. At the end of transitionToNode is the block:

if(self.onChange) {
    self.figure.selectedPath = getPath(node);
    self.onChange(self.figure);
}

So when this is called on a click, it updates the figure and we pass it back up the React chain of command. But it's also called during drawing, in which case the figure we pass back up will be the same one we just received. Which makes the next section extremely important...

Diffing

const change = diff(oldFigure, newFigure);
if(!change) { return; }

const sizeChange = change.width || change.height || change.padding;
const dataChange = change.data;

We compare the old and new figures to determine what changed. Here we're concerned with three things:

1. Are there any changes at all? If not, we can bail out now, without running any DOM manipulations. This will happen regularly due to transitionToNode as described above.
2. Did the size of the figure change? If so there are more extensive things we need to do, that will require updating the size and position of all our paths and text elements.
3. Did the data change? Inside diff we look for dataVersion, and if we find it we skip comparing data itself between the old and new figures, instead reporting changes in dataVersion as change.data.

There can be other changes that lead to a truthy change without setting either sizeChange or dataChange - such as innerRadius and selectedPath, and in general if we added styling properties (colors, line widths, font sizes...) they would fall into this category too. Those can follow the minimal update pathway below.

Drawing

if(sizeChange) { setSize(); }

let paths = self.pathGroup.selectAll('path');
let texts = self.textGroup.selectAll('text');

if(dataChange) {
    // clone data before partitioning, since this mutates the data
    self.nodes = self.partition.nodes(addIndices(JSON.parse(JSON.stringify(data))));
    paths = paths.data(self.nodes, getPathStr);
    texts = texts.data(self.nodes, getPathStr);

    // exit paths at the beginning of the transition
    // enters will happen at the end
    paths.exit().remove();
    texts.exit().remove();
}

const selectedNode = getNode(self.nodes[0], selectedPath);
// no node: path is wrong, probably because we received a new selectedPath
// before the data it belongs with
if(!selectedNode) { return retVal; }

// immediate redraw rather than transition if:
const shouldAnimate =
    // first draw
    self.initialized &&
    // new root node
    (newRootName === oldRootName) &&
    // not a pure up/down transition
    sameHead(oldSelectedPath, newSelectedPath) &&
    // the previous data didn't contain the new selected node
    // this can happen if we transition selectedPath first, then data
    (!dataChange || getNode(oldFigure.data, newSelectedPath));

console.log(shouldAnimate, oldSelectedPath, newSelectedPath);

if(shouldAnimate) {
    retVal = new Promise(resolve => {
        transitionToNode(selectedNode)
            .each('end', () => {
                updatePaths(paths, texts, dataChange);
                self.transitioning = false;
                resolve();
            });
    });
}
else {
    // first draw has no animation, and initializes the scales
    self.angularScale.domain(selectedX(selectedNode));
    self.radialScale.domain(selectedY(selectedNode))
    self.radialScale.range(selectedRadius(selectedNode));

    updatePaths(paths, texts, dataChange);

    self.initialized = true;
}

If the size and data did not change, all we do is select the paths and texts, find the selected node, transition to it, and, upon finishing that transition, update the paths - and updatePaths knows about dataChange so it can skip the enter() steps.

The logic for whether the state transition is amenable to animation or not is handled here, in shouldAnimate. This is important for Dash - and for React integration in general - because it means this is the only place we need to worry about edge detection. Dash apps are stateless, so it's particularly tricky to determine this on the Python side, and React apps are best written the same way as far down the tree as possible. D3 to a certain extent can work similarly, but for finer control we explicitly calculate what kind of change has been made and tell D3 whether to animate.

Now lets open the JavaScript demo environment:

npm run start

Lo and behold, we have a zoomable sunburst chart, connected to changing data and sibling UI controls, drawn with D3 and React 🎉 There are of course bits of polish to be added if this component were to be used in production - shrinking or removing text that's too big for its arc, and creating style props, for example, and nicer tooltips than the built-in <title> elements. But the principles are the same.

Code Walkthrough - Python side

dash-component-boilerplate makes it super easy to connect the React component we just made to Python. As in its README, run:

npm run build:js-dev
npm run build:py

For these build steps to run without warnings, the lib/components directory should contain only React components, which is why we moved the D3 code into its own directory, lib/d3. Now we can use the component in our Dash app usage.py:

from dash_sunburst import Sunburst

We'll make a simple app using this component: Feeding some static data to the component, we'll display the selected path elsewhere, and create a plotly.js graph that calculates some statistics based on the displayed data and selected path. First the static data and the app layout:

sunburst_data = { ... }

app.layout = html.Div([
    html.Div(
        [Sunburst(id='sun', data=sunburst_data)],
        style={'width': '49%', 'display': 'inline-block', 'float': 'left'}),
    dcc.Graph(
        id='graph',
        style={'width': '49%', 'display': 'inline-block', 'float': 'left'}),
    html.Div(id='output', style={'clear': 'both'})
])

Our Sunburst component doesn't support style, so we wrap it in an html.Div. The Graph and Div#output are initially blank, but our callbacks will fill them in on load. The content of these callbacks is straightforward Python - check out usage.py for the complete code - the key is simply to identify the dependencies of each one using the @app.callback decorator:

@app.callback(Output('output', 'children'), [Input('sun', 'selectedPath')])
def display_selected(selected_path):
    # format the selected path for display as text
    ...

@app.callback(Output('graph', 'figure'), [Input('sun', 'data'), Input('sun', 'selectedPath')])
def display_graph(data, selected_path):
    # crawl the sunburst data, along with its selected path,
    # to create the related plotly.js graph
    ...

And that's it! python usage.py gives us our D3 sunburst diagram, connected through Dash to whatever else we choose.

Further examples expanding on server-side updates can be found in usage_backend_update_via_controls.py and usage_backend_update_via_selections.py

More Resources

• Learn more about Dash: https://dash.plot.ly
• View the original component boilerplate: https://github.com/plotly/dash-component-boilerplate

dash-sunburst's Issues

No pulldown menu

Hi,

I'm testing this out and noticed that the pulldown menu that was shown in the screenshots for selecting the levels to plot is no longer visible. The page has to be reloaded if you want to switch to a different level.

Add `maxLevels` prop for the number of levels to display

If you have data with highly nested nodes, it might be preferable to only show a few levels at a time, diving deeper as you select inner nodes.

If the data were TOO big (so performance would lag if you send it all to the front end) you need to do something like usage_backend_update_via_selections to fetch the appropriate data whenever the selection changes. But for intermediate-size data, we should be able to simply restrict the number of visible levels by adding a prop called something like maxLevels.

The key to this will be altering the radial range, which is set by selectedY

d3.layout.partition (note we're using v3 here) sets node.y and node.dy for all nodes such that the root node has y=0 and the deepest level of nesting has y+dy=1, while dy is 1/(total number of levels)

So the main thing we need is to change selectedY from:

node => [node.y, 1]

to something like:

node => [node.y,  maxLevels ? Math.min(1, node.y + maxLevels * node.dy)] : 1]

ie if maxLevels is 0 or missing, keep the current behavior, otherwise restrict the end of the radial range to no more than the requested number.

Then maxLevels needs to be propagated up from the d3 component to the react wrapper component (the d3 component receives props from the react component as figure here)

It's possible that other things will be needed, like explicitly removing text and/or path elements that are collapsed against the outside edge... we'll have to see how it looks and performs to determine if this is needed.

Value of 'values' is not the name of a column in 'data_frame'. Expected one of [0] but received column name

Hi

I am working in Dash app,
I am rendering express sunburst chart, with "pyspark.pandas" dataframe, but I am getting below error in dash.
ERROR : Value of 'values' is not the name of a column in 'data_frame'. Expected one of [0] but received: column_3

I have provided data in sunburst chart as below:
px.sunburst(pdf, path=["column_1", "column_2"], values="column_3")

Here column_3 is a integer.

Please help me, how can I resolve this issue.

Thanks
Nagaraja M M