WebGl 2D Scatterplot with Regl

A highly-scalable pan-and-zoomable scatter plot library that uses WebGL through Regl. This library sacrifices feature richness for speed to allow rendering up to 20 million points (depending on your hardware of course) including fast lasso selection. Further, the footprint of regl-scatterplot is kept small. NEW: Python lovers please see jscatter: a Jupyter Notebook/Lab widget that uses regl-scatterplot.

Demo: https://flekschas.github.io/regl-scatterplot/

Live Playground: https://observablehq.com/@flekschas/regl-scatterplot

Default Interactions:

• Pan: Click and drag your mouse.

• Zoom: Scroll vertically.

• Rotate: While pressing ALT, click and drag your mouse.

• Select a dot: Click on a dot with your mouse.

• Select multiple dots:

  • While pressing SHIFT, click and drag your mouse. All items within the lasso will be selected.

  • Upon activating the lasso initiator (i.e., lassoInitiator: true) you can click into the background and a circle will appear under your mouse cursor. Click inside this circle and drag your mouse to start lassoing.

    Click here to see how it works

    

• Deselect: Double-click onto an empty region.

Note, you can remap rotate and lasso to other modifier keys via the keyMap option!

Supported Visual Encodings:

• x/y point position (obviously)
• categorical and continuous color encoding (including opacity)
• categorical and continuous size encoding
• point connections (stemming, for example, from time series data)

Install

npm i regl-scatterplot

FYI, if you're using npm version prior to 7, you have to install regl-scatterplot's peer dependencies (regl and pub-sub-es) manually.

Getting started

Basic Example

import createScatterplot from 'regl-scatterplot';

const canvas = document.querySelector('#canvas');

const { width, height } = canvas.getBoundingClientRect();

const scatterplot = createScatterplot({
  canvas,
  width,
  height,
  pointSize: 5,
});

const points = new Array(10000)
  .fill()
  .map(() => [-1 + Math.random() * 2, -1 + Math.random() * 2, color]);

scatterplot.draw(points);

IMPORTANT: Your points positions need to be normalized to [-1, 1] (normalized device coordinates). Why? Regl-scatterplot is designed to be a lower-level library, whose primary purpose is speed. As such it expects you to normalize the data upfront.

Color, Opacity, and Size Encoding

In regl-scatterplot, points can be associated with two data values. These two values are defined as the third and forth component of the point quadruples ([x, y, value, value]). For instance:

scatterplot.draw([
  [0.2, -0.1, 0, 0.1337],
  [0.3, 0.1, 1, 0.3371],
  [-0.9, 0.8, 2, 0.3713],
]);

These two values can be visually encoded as the color, opacity, or the size. Integers are treated as categorical data and floats that range between [0, 1] are treated as continuous values. In the example above, the first point value would be treated as categorical data and the second would be treated as continuous data.

In the edge case that you have continuous data but all data points are either 0 or 1 you can manually set the data type via the zDataType and wDatatype draw options.

To encode the two point values use the colorBy, opacityBy, and sizeBy property as follows:

scatterplot.set({
  opacityBy: 'valueA',
  sizeBy: 'valueA',
  colorBy: 'valueB',
});

In this example we would encode the first categorical point values ([0, 1, 2]) as the point opacity and size. The second continuous point values ([0.1337, 0.3317, 0.3713]) would be encoded as the point color.

The last thing we need to tell regl-scatterplot is what those point values should be translated to. We do this by specifying a color, opacity, and size map as an array of colors, opacities, and sizes as follows:

scatterplot.set({
  pointColor: ['#000000', '#111111', ..., '#eeeeee', '#ffffff'],
  pointSize: [2, 4, 8],
  opacity: [0.5, 0.75, 1],
});

You can encode a point data value in multiple ways. For instance, as you can see in the example above, the categorical fist data value is encoded via the point size and opacity.

What if I have more than two values associated to a point? Unfortunately, this isn't supported currently. In case you're wondering, this limitation is due to how we store the point data. The whole point state is encoded as an RGBA texture where the x and y coordinate are stored as the red and green color components and the first and second data value are stored in the blue and alpha component of the color. However, this limitation might be addressed in future versions so make sure to check back or, even better, start a pull request!

Why can't I specify a range function instead of a map? Until we have implemented enough scale functions in the shader it's easier to let you pre-compute the map. For instance, if you wanted to encode a continuous values on a log scale of point size, you can simply do pointSize: Array(100).fill().map((v, i) => Math.log(i + 1) + 1).

Code Example | Demo

Connecting points

You can connect points visually using spline curves by adding a 5th component to your point data and setting showPointConnections: true.

The 5th component is needed to identify which points should be connected. By default, the order of how the points are connected is defined by the order in which the points appear in your data.

const points = [
  [1, 1, 0, 0, 0],
  [2, 2, 0, 0, 0],
  [3, 3, 0, 0, 1],
  [4, 4, 0, 0, 1],
  [5, 5, 0, 0, 0],
];

In the example above, the points would be connected as follows:

0 -> 1 -> 4
2 -> 3

Line Ordering:

To explicitely define or change the order of how points are connected, you can define a 6th component as follows:

const points = [
  [1, 1, 0, 0, 0, 2],
  [2, 2, 0, 0, 0, 0],
  [3, 3, 0, 0, 1, 1],
  [4, 4, 0, 0, 1, 0],
  [5, 5, 0, 0, 0, 1],
];

would lead tp the following line segment ordering:

1 -> 4 -> 0
3 -> 2

Note, to visualize the point connections, make sure scatterplot.set({ showPointConnection: true }) is set!

Code Example | Demo

Synchronize D3 x and y scales with the scatterplot view

Under the hood regl-scatterplot uses a 2D camera, which you can either get via scatterplot.get('camera') or scatterplot.subscribe('view', ({ camera }) => {}). You can use the camera's view matrix to compute the x and y scale domains. However, since this is tedious, regl-scatterplot allows you to specify D3 x and y scales that will automatically be synchronized. For example:

const xScale = scaleLinear().domain([0, 42]);
const yScale = scaleLinear().domain([-5, 5]);
const scatterplot = createScatterplot({
  canvas,
  width,
  height,
  xScale,
  yScale,
});

Now whenever you pan or zoom, the domains of xScale and yScale will be updated according to your current view. Note, the ranges are automatically set to the width and height of your canvas object.

Code Example | Demo

Translating Point Coordinates to Screen Coordinates

Imagine you want to render additional features on top of points points, for which you need to know where on the canvas points are drawn. To determine the screen coordinates of points you can use D3 scales and scatterplot.get('pointsInView') as follows:

const points = Array.from({ length: 100 }, () => [Math.random() * 42, Math.random()]);
const [xScale, yScale] = [scaleLinear().domain([0, 42]), scaleLinear().domain([0, 1])];

const scatterplot = createScatterplot({ ..., xScale, yScale });
scatterplot.draw(points);

scatterplot.subscribe('view', ({ xScale, yScale }) => {
  console.log('pointsInScreenCoords', scatterplot.get('pointsInView').map((pointIndex) => [
    xScale(points[pointIndex][0]),
    yScale(points[pointIndex][1])
  ]));
});

Code Example | Demo

Transition Points

To make sense of two different states of points, it can help to show an animation by transitioning the points from their first to their second location. To do so, simple draw() the new points as follows:

const initialPoints = Array.from({ length: 100 }, () => [Math.random() * 42, Math.random()]);
const finalPoints = Array.from({ length: 100 }, () => [Math.random() * 42, Math.random()]);

const scatterplot = createScatterplot({ ... });
scatterplot.draw(initialPoints).then(() => {
  scatterplot.draw(finalPoints, { transition: true });
})

It's important that the number of points is the same for the two draw() calls. Also note that the point correspondence is determined by their index.

Code Example | Demo

Zoom to Points

Sometimes it can be useful to programmatically zoom to a set of points. In regl-scatterplot you can do this with the zoomToPoints() method as follows:

const points = Array.from({ length: 100 }, () => [Math.random() * 42, Math.random()]);

const scatterplot = createScatterplot({ ... });
scatterplot.draw(initialPoints).then(() => {
  // We'll select the first five points...
  scatterplot.select([0, 1, 2, 3, 4]);
  // ...and zoom into them
  scatterplot.zoomToPoints([0, 1, 2, 3, 4], { transition: true })
})

Note that the zooming can be smoothly transitioned when { transition: true } is passed to the function.

Code Example | Demo

API

Constructors

# createScatterplot(options = {})

Returns: a new scatterplot instance.

Options: is an object that accepts any of the properties.

# createRenderer(options = {})

Returns: a new Renderer instance with appropriate extensions being enabled.

Options: is an object that accepts any of the following optional properties:

• regl: a Regl instance to be used for rendering.
• canvas: background color of the scatterplot.
• gamma: the gamma value for alpha blending.

# createRegl(canvas)

Returns: a new Regl instance with appropriate extensions being enabled.

Canvas: the canvas object on which the scatterplot will be rendered on.

# createTextureFromUrl(regl, url)

DEPRECATED! Use scatterplot.createTextureFromUrl() instead.

Methods

# scatterplot.draw(points, options)

Sets and draws points. Importantly, the points' x and y coordinates need to have been normalized to [-1, 1] (normalized device coordinates). The two additional values (valueA and valueB) need to be normalized to [0, 1] (if they represent continuous data) or [0, >1] (if they represent categorical data).

Note that repeatedly calling this method without specifying points will not clear previously set points. To clear points use scatterplot.clear().

Arguments:

• points can either be an array of quadruples (row-oriented) or an object of arrays (column-oriented):
  • For row-oriented data, each nested array defines a point data of the form [x, y, ?valueA, ?valueB, ?line, ?lineOrder]. valueA and valueB are optional and can be used for color, opacity, or size encoding. line and lineOrder are also optional and can be used to visually connect points by lines.
  • For column-oriented data, the object must be of the form { x: [], y: [], ?valueA: [], ?valueB: [], ?line: [], ?lineOrder: [] }.
• options is an object with the following properties:
  • showPointConnectionsOnce [default: false]: if true and if points contain a line component/dimension the points will be visually conntected.
  • transition [default: false]: if true and if the current number of points equals points.length, the current points will be transitioned to the new points
  • transitionDuration [default: 500]: the duration in milliseconds over which the transition should occur
  • transitionEasing [default: cubicInOut]: the easing function, which determines how intermediate values of the transition are calculated
  • preventFilterReset [default: false]: if true and if the number of new points is the same as the current number of points, the current point filter will not be reset
  • hover [default: undefined]: a shortcut for hover(). This option allows to programmatically hover a point by specifying a point index
  • select [default: undefined]: a shortcut for select(). This option allows to programmatically select points by specifying a list of point indices
  • filter [default: undefined]: a shortcut for filter(). This option allows to programmatically filter points by specifying a list of point indices
  • zDataType [default: undefined]: This option allows to manually set the data type of the z/valueA value to either continuous or categorical. By default the data type is determined automatically.
  • wDataType [default: undefined]: This option allows to manually set the data type of the w/valyeB value to either continuous or categorical. By default the data type is determined automatically.

Returns: a Promise object that resolves once the points have been drawn or transitioned.

Examples:

const points = [
  [
    // The relative X position in [-1,1] (normalized device coordinates)
    0.9,
    // The relative Y position in [-1,1] (normalized device coordinates)
    0.3,
    // The category, which defaults to `0` if `undefined`
    0,
    // A continuous value between [0,1], which defaults to `0` if `undefined`
    0.5,
  ],
];

scatterplot.draw(points);

// You can now do something else like changing the point size etc.

// If we want to animate the transition of our point from above to another
// x,y position, we can also do this by drawing a new point while enableing
// transition via the `options` argument.
scatterplot.draw([[0.6, 0.6, 0, 0.6]], { transition: true });

// Let's unset the points. To do so, pass in an empty array to `draw()`.
// Or alternatively, call `scatterplot.clear()`
scatterplot.draw([]);

// You can also specify the point data in a column-oriented format. The
// following call will draw three points: (1,3), (2,2), and (3,1)
scatterplot.draw({
  x: [1, 2, 3],
  y: [3, 2, 1],
});

// Finally, you can also specify which point will be hovered, which points will
// be selected, and which points will be filtered. These options are useful to
// avoid a flicker which would occur if `hover()`, `select()`, and `filter()`
// are called after `draw()`.
scatterplot.draw(
  { x: [1, 2, 3], y: [3, 2, 1] },
  { hover: 0, selected: [0, 1], filter: [0, 2] }
);

# scatterplot.redraw()

Redraw the scatter plot at the next animation frame.

Note, that regl-scatterlot automatically redraws the scatter plot whenever the view changes in some ways. So theoretically, there should never be a need to call this function!

# scatterplot.clear()

Clears previously drawn points.

# scatterplot.get(property)

Arguments:

• property is a string referencing a property.

Returns: the property value.

# scatterplot.set(properties = {})

Arguments:

• properties is an object of key-value pairs. See below for a list of all properties.

# scatterplot.select(points, options = {})

Select some points, such that they get visually highlighted. This will trigger a select event unless options.preventEvent === true.

Arguments:

• points is an array of point indices referencing the points that you want to select.
• options [optional] is an object with the following properties:
  • preventEvent: if true the select will not be published.

Examples:

// Let's say we have three points
scatterplot.draw([
  [0.1, 0.1],
  [0.2, 0.2],
  [0.3, 0.3],
]);

// To select the first and second point we have to do
scatterplot.select([0, 1]);

# scatterplot.deselect(options = {})

Deselect all selected points. This will trigger a deselect event unless options.preventEvent === true.

# scatterplot.filter(points, options = {})

Filter down the currently drawn points, such that all points that are not included in the filter are visually and interactivelly hidden. This will trigger a filter event unless options.preventEvent === true.

Note: filtering down points can affect previously selected points. Selected points that are filtered out are also deselected.

Arguments:

• points is an array of indices referencing the points that you want to filter down to.
• options [optional] is an object with the following properties:
  • preventEvent: if true the select will not be published.

Examples:

// Let's say we have three points
scatterplot.draw([
  [0.1, 0.1],
  [0.2, 0.2],
  [0.3, 0.3],
]);

// To only show the first and second point we have to do
scatterplot.filter([0, 1]);

# scatterplot.unfilter(options = {})

Reset previously filtered out points. This will trigger an unfilter event unless options.preventEvent === true.

# scatterplot.hover(point, options = {})

Programmatically hover a point, such that it gets visually highlighted. This will trigger a pointover or pointout event unless options.preventEvent === true.

Arguments:

• point is the point index referring to the point you want to hover.
• options [optional] is an object with the following properties:
  • showReticleOnce: if true the reticle will be shown once, even if showReticle === false.
  • preventEvent: if true the pointover and pointout will not be published.

Examples:

scatterplot.draw([
  [0.1, 0.1],
  [0.2, 0.2],
  [0.3, 0.3],
]);

scatterplot.hover(1); // To hover the second point

Arguments:

• options [optional] is an object with the following properties:
• preventEvent: if true the deselect will not be published.

# scatterplot.zoomToPoints(points, options = {})

Zoom to a set of points

Arguments:

• points is an array of point indices.
• options [optional] is an object with the following properties:
  • padding: [default: 0]: relative padding around the bounding box of the points to zoom to
  • transition [default: false]: if true, the camera will smoothly transition to its new position
  • transitionDuration [default: 500]: the duration in milliseconds over which the transition should occur
  • transitionEasing [default: cubicInOut]: the easing function, which determines how intermediate values of the transition are calculated

Examples:

// Let's say we have three points
scatterplot.draw([
  [0.1, 0.1],
  [0.2, 0.2],
  [0.3, 0.3],
]);

// To zoom to the first and second point we have to do
scatterplot.zoomToPoints([0, 1]);

# scatterplot.zoomToOrigin(options = {})

Zoom to the original camera position. This is similar to resetting the view

Arguments:

• options [optional] is an object with the following properties:
  • transition [default: false]: if true, the camera will smoothly transition to its new position
  • transitionDuration [default: 500]: the duration in milliseconds over which the transition should occur
  • transitionEasing [default: cubicInOut]: the easing function, which determines how intermediate values of the transition are calculated

# scatterplot.zoomToLocation(target, distance, options = {})

Zoom to a specific location, specified in normalized device coordinates. This function is similar to scatterplot.lookAt(), however, it allows to smoothly transition the camera position.

Arguments:

• target the camera target given as a [x, y] tuple.
• distance the camera distance to the target given as a number between ]0, Infinity]. The smaller the number the closer moves the camera, i.e., the more the view is zoomed in.
• options [optional] is an object with the following properties:
  • transition [default: false]: if true, the camera will smoothly transition to its new position
  • transitionDuration [default: 500]: the duration in milliseconds over which the transition should occur
  • transitionEasing [default: cubicInOut]: the easing function, which determines how intermediate values of the transition are calculated

Examples:

scatterplot.zoomToLocation([0.5, 0.5], 0.5, { transition: true });
// => This will make the camera zoom into the top-right corner of the scatter plot

# scatterplot.zoomToArea(rectangle, options = {})

Zoom to a specific area specified by a recangle in normalized device coordinates.

Arguments:

• rectangle the rectangle must come in the form of { x, y, width, height }.
• options [optional] is an object with the following properties:
  • transition [default: false]: if true, the camera will smoothly transition to its new position
  • transitionDuration [default: 500]: the duration in milliseconds over which the transition should occur
  • transitionEasing [default: cubicInOut]: the easing function, which determines how intermediate values of the transition are calculated
  • preventFilterReset [default: false]: if true and if the number of new points equals the number of already drawn points, the point filter set is not being reset.

Examples:

scatterplot.zoomToArea(
  { x: 0, y: 0, width: 1, height: 1 },
  { transition: true }
);
// => This will make the camera zoom into the top-right corner of the scatter plot

# scatterplot.getScreenPosition(pointIdx)

Get the screen position of a point

Arguments:

• pointIdx is a point indix.

Examples:

// Let's say we have a 100x100 pixel scatter plot with three points
const scatterplot = createScatterplot({ width: 100, height: 100 });
scatterplot.draw([
  [-1, -1],
  [0, 0],
  [1, 1],
]);

// To retrieve the screen position of the second point you can call. If we
// haven't panned and zoomed, the returned position should be `50, 50`
scatterplot.getScreenPosition(1);
// => [50, 50]

# scatterplot.lookAt(view, options = {})

Update the camera's view matrix to change the viewport. This will trigger a view event unless options.preventEvent === true.

Note, this API is a shorthand to scatterplot.set({ 'cameraView': view }) with the additional features of allowing to prevent view events.

# scatterplot.destroy()

Destroys the scatterplot instance by disposing all event listeners, the pubSub instance, regl, and the camera.

# scatterplot.refresh()

Refreshes the viewport of the scatterplot's regl instance.

# scatterplot.reset(options)

Sets the view back to the initially defined view. This will trigger a view event unless options.preventEvent === true.

# scatterplot.export(options)

Arguments:

• options is an object for customizing how to export. See regl.read() for details.

Returns: an object with three properties: pixels, width, and height. The pixels is a Uint8ClampedArray.

# scatterplot.subscribe(eventName, eventHandler)

Subscribe to an event.

Arguments:

• eventName needs to be a valid event name.
• eventHandler needs to be a callback function that can receive the payload.

Returns: an unsubscriber object that can be passed into unsubscribe().

# scatterplot.unsubscribe(eventName, eventHandler)

Unsubscribe from an event. See scatterplot.subscribe() for a list of all events.

# scatterplot.createTextureFromUrl(url)

Returns: a Promise that resolves to a Regl texture that can be used, for example, as the background image.

url: the URL to an image.

Properties

You can customize the scatter plot according to the following properties that can be read and written via scatterplot.get() and scatterplot.set().

Name	Type	Default	Constraints	Settable	Nullifiable
canvas	object	document.createElement('canvas')		false	false
regl	Regl	createRegl(canvas)		false	false
renderer	Renderer	createRenderer()		false	false
syncEvents	boolean	false		false	false
version	string			false	false
width	int or str	'auto'	'auto' or > 0	true	false
height	int or str	'auto'	'auto' or > 0	true	false
aspectRatio	float	1.0	> 0	true	false
backgroundColor	string or array	rgba(0, 0, 0, 1)	hex, rgb, rgba	true	false
backgroundImage	function	null	Regl texture	true	true
camera	object		See dom-2d-camera	false	false
cameraTarget	tuple	[0, 0]		true	false
cameraDistance	float	1	> 0	true	false
cameraRotation	float	0		true	false
cameraView	Float32Array	[1,0,0,0,0,1,0,0,0,0,1,0,0,0,0,1]		true	false
colorBy	string	null	See data encoding	true	true
sizeBy	string	null	See data encoding	true	true
opacityBy	string	null	See data encoding	true	true
deselectOnDblClick	boolean	true		true	false
deselectOnEscape	boolean	true		true	false
opacity	float	1	Must be in ]0, 1]	true	false
opacityInactiveMax	float	1	Must be in [0, 1]	true	false
opacityInactiveScale	float	1	Must be in [0, 1]	true	false
points	tuple[]	[[0.5, 2.3], ...]		false	false
selectedPoints	int[]	[4, 2]		false	false
filteredPoints	int[]	[4, 2]		false	false
pointsInView	int[]	[1, 2, 12]		false	false
pointColor	quadruple	[0.66, 0.66, 0.66, 1]	single value or list of hex, rgb, rgba	true	false
pointColorActive	quadruple	[0, 0.55, 1, 1]	single value or list of hex, rgb, rgba	true	false
pointColorHover	quadruple	[1, 1, 1, 1]	single value or list of hex, rgb, rgba	true	false
pointOutlineWidth	int	2	>= 0	true	false
pointSize	int	6	> 0	true	false
pointSizeSelected	int	2	>= 0	true	false
showPointConnection	boolean	false		true	false
pointConnectionColor	quadruple	[0.66, 0.66, 0.66, 0.2]		true	false
pointConnectionColorActive	quadruple	[0, 0.55, 1, 1]		true	false
pointConnectionColorHover	quadruple	[1, 1, 1, 1]		true	false
pointConnectionColorBy	string	null	See data encoding	true	false
pointConnectionOpacity	float	0.1		true	false
pointConnectionOpacityActive	float	0.66		true	false
pointConnectionOpacityBy	string	null	See data encoding	true	false
pointConnectionSize	float	2		true	false
pointConnectionSizeActive	float	2		true	false
pointConnectionSizeBy	string	null	See data encoding	true	false
pointConnectionMaxIntPointsPerSegment	int	100		true	false
pointConnectionTolerance	float	0.002		true	false
lassoColor	quadruple	rgba(0, 0.667, 1, 1)	hex, rgb, rgba	true	false
lassoLineWidth	float	2	>= 1	true	false
lassoMinDelay	int	15	>= 0	true	false
lassoMinDist	int	4	>= 0	true	false
lassoClearEvent	string	'lassoEnd'	'lassoEnd' or 'deselect'	true	false
lassoInitiator	boolean	false		true	false
lassoInitiatorElement	object	the lasso dom element		false	false
lassoInitiatorParentElement	object	document.body		true	false
showReticle	boolean	false	true or false	true	false
reticleColor	quadruple	rgba(1, 1, 1, .5)	hex, rgb, rgba	true	false
xScale	function	null	must follow the D3 scale API	true	true
yScale	function	null	must follow the D3 scale API	true	true
keyMap	object	{ alt: 'rotate', shift: 'lasso' }	See the notes below	true	false
mouseMode	string	'panZoom'	'panZoom', 'lasso', or 'rotate'	true	false
performanceMode	boolean	false	can only be set during initialization!	true	false
gamma	float	1	to control the opacity blending	true	false
isDestroyed	boolean	false		false	false
isPointsDrawn	boolean	false		false	false
isPointsFiltered	boolean	false		false	false

# Notes:

• An attribute is considered nullifiable if it can be unset. Attributes that are not nullifiable will be ignored if you try to set them to a falsy value. For example, if you call scatterplot.attr({ width: 0 }); the width will not be changed as 0 is interpreted as a falsy value.

• By default, the width and height are set to 'auto', which will make the canvas stretch all the way to the bounds of its clostest parent element with position: relative. When set to 'auto' the library also takes care of resizing the canvas on resize and orientationchange events.

• The background of the scatterplot is transparent, i.e., you have to control the background with CSS! background is used when drawing the outline of selected points to simulate the padded border only.

• The background image must be a Regl texture. To easily set a remote image as the background please use createTextureFromUrl.

• The scatterplot understan 4 colors per color representing 4 states, representing:

  • normal (pointColor): the normal color of points.
  • active (pointColorActive): used for coloring selected points.
  • hover (pointColorHover): used when mousing over a point.
  • background (backgroundColor): used as the background color.

• Points can currently by colored by category and value.

• The size of selected points is given by pointSize + pointSizeSelected

• By default, events are published asynchronously to decouple regl-scatterplot's execution flow from the event consumer's process. However, you can enable synchronous event broadcasting at your own risk via createScatterplot({ syncEvents: true }). This property can't be changed after initialization!

• If you need to draw more than 2 million points, you might want to set performanceMode to true during the initialization to boost the performance. In performance mode, points will be drawn as simple squares and color blending is disabled. This should allow you to draw up to 20 million points (or more depending on your hardware). Make sure to reduce the pointSize as you render more and more points (e.g., 0.25 for 20 million works for me) to ensure good performance.

# colorBy, opacityBy, sizeBy:

To visual encode one of the two point values set colorBy, opacityBy, or sizeBy to one of the following values referencing the third or forth component of your points. To reference the third component you can use category (only for backwards compatibility), value1, valueA, valueZ, or z. To reference the forth component use value (only for backwards compatibility), value2, valueB, valueW, or w.

Density-based opacity encoding: In addition, the opacity can dynamically be set based on the point density and zoom level via opacityBy: 'density'. As an example go to dynamic-opacity.html. The implementation is an extension of Ricky Reusser's awesome notebook. Huuuge kudos Ricky! 🙇‍♂️

# pointConnectionColorBy, pointConnectionOpacityBy, and pointConnectionSizeBy:

In addition to the properties understood by colorBy, etc., pointConnectionColorBy, pointConnectionOpacityBy, and pointConnectionSizeBy also understand "inherit" and "segment". When set to "inherit", the value will be inherited from its point-specific counterpart. When set to "segment", each segment of a point connection will be encoded separately. This allows you to, for instance, color connection by a gradient from the start to the end of each line.

# lassoInitiator:

When setting lassoInitiator to true you can initiate the lasso selection without the need to hold down a modifier key. Simply click somewhere into the background and a circle will appear under your mouse cursor. Now click into the circle and drag you mouse to start lassoing. You can additionally invoke the lasso initiator circle by a long click on a dot.

You don't like the look of the lasso initiator? No problem. Simple get the DOM element via scatterplot.get('lassoInitiatorElement') and adjust the style via JavaScript. E.g.: scatterplot.get('lassoInitiatorElement').style.background = 'green'.

# KeyMap:

The keyMap property is an object defining which actions are enabled when holding down which modifier key. E.g.: { shift: 'lasso' }. Acceptable modifier keys are alt, cmd, ctrl, meta, shift. Acceptable actions are lasso, rotate, and merge (for selecting multiple items by merging a series of lasso or click selections).

You can also use the keyMap option to disable the lasso selection and rotation by setting keyMap to an empty object.

# Examples:

// Set width and height
scatterplot.set({ width: 300, height: 200 });

// get width
const width = scatterplot.get('width');

// Set the aspect ratio of the scatterplot. This aspect ratio is referring to
// your data source and **not** the aspect ratio of the canvas element! By
// default it is assumed that your data us following a 1:1 ratio and this ratio
// is preserved even if your canvas element has some other aspect ratio. But if
// you wanted you could provide data that's going from [0,2] in x and [0,1] in y
// in which case you'd have to set the aspect ratio as follows to `2`.
scatterplot.set({ aspectRatio: 2.0 });

// Set background color to red
scatterplot.set({ backgroundColor: '#00ff00' }); // hex string
scatterplot.set({ backgroundColor: [255, 0, 0] }); // rgb array
scatterplot.set({ backgroundColor: [255, 0, 0, 1.0] }); // rgba array
scatterplot.set({ backgroundColor: [1.0, 0, 0, 1.0] }); // normalized rgba

// Set background image to an image
scatterplot.set({ backgroundImage: 'https://server.com/my-image.png' });
// If you need to know when the image was loaded you have two options. First,
// you can listen to the following event
scatterplot.subscribe(
  'backgroundImageReady',
  () => {
    console.log('Background image is now loaded and rendered!');
  },
  1
);
// or you load the image yourself as follows
const backgroundImage = await scatterplot.createTextureFromUrl(
  'https://server.com/my-image.png'
);
scatterplot.set({ backgroundImage });

// Color by
scatterplot.set({ colorBy: 'category' });

// Set color map
scatterplot.set({
  pointColor: ['#ff0000', '#00ff00', '#0000ff'],
  pointColorActive: ['#ff0000', '#00ff00', '#0000ff'], // optional
  pointColorHover: ['#ff0000', '#00ff00', '#0000ff'], // optional
});

// Set base opacity
scatterplot.set({ opacity: 0.5 });

// If you want to deemphasize unselected points (when some points are selected)
// you can rescale the unselected points' opacity as follows
scatterplot.set({ opacityInactiveScale: 0.5 });

// Set the width of the outline of selected points
scatterplot.set({ pointOutlineWidth: 2 });

// Set the base point size
scatterplot.set({ pointSize: 10 });

// Set the additional point size of selected points
scatterplot.set({ pointSizeSelected: 2 });

// Change the lasso color and make it very smooth, i.e., do not wait before
// extending the lasso (i.e., `lassoMinDelay = 0`) and extend the lasso when
// the mouse moves at least 1 pixel
scatterplot.set({
  lassoColor: [1, 1, 1, 1],
  lassoMinDelay: 0,
  lassoMinDist: 1,
  // This will keep the drawn lasso until the selected points are deselected
  lassoClearEvent: 'deselect',
});

// Activate reticle and set reticle color to red
scatterplot.set({ showReticle: true, reticleColor: [1, 0, 0, 0.66] });

Renderer

The renderer class is responsible for rendering pixels onto the scatter plot's canvas using WebGL via Regl. It's created automatically internally but you can also create it yourself, which can be useful when you want to instantiate multiple scatter plot instances as they can share one renderer.

Renderer API

# renderer.canvas

The renderer's canvas instance. (Read-only)

# renderer.gamma

The renderer's gamma value. This value influences the alpha blending.

# renderer.regl

The renderer's regl instance. (Read-only)

# renderer.onFrame(function)

Add a function to be called on every animation frame.

Arguments:

• function: The function to be called on every animation frame.

Returns: A function to remove the added function from the animation frame cycle.

# renderer.refresh()

Updates Regl's viewport, drawingBufferWidth, and drawingBufferHeight.

# renderer.render(drawFunction, targetCanvas)

Render Regl draw instructions into a target canvas using the renderer.

Arguments:

• drawFunction: The draw function that triggers Regl draw instructions
• targetCanvas: The canvas to rendering the final pixels into.

Events

Name	Trigger	Payload
init	when the scatter plot is initialized	undefined
destroy	when the scatter plot is destroyed	undefined
backgroundImageReady	when the background image was loaded	undefined
pointOver	when the mouse cursor is over a point	pointIndex
pointOut	when the mouse cursor moves out of a point	pointIndex
select	when points are selected	{ points }
deselect	when points are deselected	undefined
filter	when points are filtered	{ points }
unfilter	when the point filter is reset	undefined
view	when the view has changes	{ camera, view, xScale, yScale }
draw	when the plot was drawn	{ camera, view, xScale, yScale }
drawing	when the plot is being drawn	{ camera, view, xScale, yScale }
lassoStart	when the lasso selection has started	undefined
lassoExtend	when the lasso selection has extended	{ coordinates }
lassoEnd	when the lasso selection has ended	{ coordinates }
transitionStart	when points started to transition	undefined
transitionEnd	when points ended to transition	createRegl(canvas)
pointConnectionsDraw	when point connections were drawn	undefined

Trouble Shooting

Resizing the scatterplot

The chances are high that you use the regl-scatterplot in a dynamically-resizable or interactive web-app. Please note that regl-scatterplot doesn't not automatically resize when the dimensions of its parent container change. It's your job to keep the size of regl-scatterplot and its parent element in sync. Hence, every time the size of the parent or canvas element changed, you have to call:

const { width, height } = canvas.getBoundingClientRect();
scatterplot.set({ width, height });

Using regl-scatterplot with Vue

Related to the resizing, when conditionally displaying regl-scatterplot in Vue you might have to update the width and height when the visibility is changed. See issue #20 for an example.

Citation

If you like regl-scatterplot and are using it in your research, we'd appreciate if you could cite our paper:

@article {lekschas2023reglscatterplot,
  author = {Lekschas, Fritz},
  title = {Regl-Scatterplot: A Scalable Interactive JavaScript-based Scatter Plot Library},
  journal = {Journal of Open Source Software},
  volume = {8},
  number = {84},
  pages = {5275},
  year = {2023},
  month = {4},
  doi = {10.21105/joss.05275},
  url = {https://doi.org/10.21105/joss.05275},
}