Skip to content

# Transforms â€‹

Transforms derive data as part of the plot specification. This accelerates what is often the most onerous task in visualization: getting data into the right shape.

For example, given a dataset of highway traffic measured as vehicles per hour by location, plotting every observation is straightforward: use a tick (or dot) and assign x = vehicles per hour and y = location. But to draw a quantifiable insight, we may want a summary statistic such as the median traffic by location. đź‘©â€Ťđź’» Below we use the group transform to group by location and apply a median reducer to position the red tick.

Fork
js
``````Plot.plot({
marginLeft: 120,
x: {label: "Vehicles per hour (thousands)", transform: (x) => x / 1000},
y: {label: null},
marks: [
Plot.ruleX([0]),
Plot.tickX(
traffic,
{x: "vehicles", y: "location", strokeOpacity: 0.3}
),
Plot.tickX(
traffic,
Plot.groupY(
{x: "median"},
{x: "vehicles", y: "location", stroke: "red", strokeWidth: 4, sort: {y: "x"}}
)
)
]
})``````

As you might expect, traffic varies significantly throughout the day, so perhaps it would be better to look at the median by hour by location? Instead of grouping only by y, we can group by both x and y to produce a heatmap.

Fork
js
``````Plot.plot({
marginLeft: 120,
padding: 0,
y: {label: null},
color: {scheme: "YlGnBu", legend: true, zero: true},
marks: [
Plot.cell(
traffic,
Plot.group(
{fill: "median"},
{x: (d) => d.date.getUTCHours(), y: "location", fill: "vehicles", inset: 0.5, sort: {y: "fill"}}
)
)
]
})``````

Plot includes many useful transforms! For example, you can compute a rolling average to smooth a noisy signal, stack layers for a streamgraph, or dodge dots for a beeswarm. Plotâ€™s various built-in transforms include: bin, centroid, dodge, filter, group, hexbin, interval, map, normalize, reverse, select, shuffle, sort, stack, tree, and window. If these donâ€™t meet your needs, you can even implement a custom transform.

Transforms are never required â€”Â you can always aggregate and derive data yourself outside of Plot, and then pass in the binned values. For example, we could use d3.bin to compute a histogram of athletesâ€™Â weights as an array of {x0, x1, length} objects.

js
``bins = d3.bin().thresholds(80).value((d) => d.weight)(olympians)``

We can then pass that to the rect mark, assigning to the x1, x2, and y2 channels:

js
``Plot.rectY(bins, {x1: "x0", x2: "x1", y2: "length"}).plot()``

INFO

This is for demonstration only; you wouldnâ€™t normally bin â€śby handâ€ťÂ as shown here.

But Plotâ€™s transforms are often more convenient, especially in conjunction with Plotâ€™s other features such as faceting and automatic grouping by z. For example, if we want to add a color encoding to our histogram, we simply add the fill option and the bin transform partitions each bin accordingly; doing this with d3.bin would be a lot more work!

Fork
js
``Plot.rectY(olympians, Plot.binX({y: "count"}, {x: "weight", fill: "sex"})).plot({color: {legend: true}})``

Plotâ€™s transforms typically take two options objects as arguments: the first object contains the transform options (above, `{y: "count"}`), while the second object contains mark options to be â€śpassed throughâ€ť to the mark (`{x: "weight", fill: "sex"}`). The transform returns a new options object representing the transformed mark options to be passed to a mark.

Breaking down the above code:

js
``````const options = {x: "weight", fill: "sex"}; // initial mark options
const binOptions = {y: "count"}; // bin transform options
const binned = Plot.binX(binOptions, options); // transformed mark options
const rect = Plot.rectY(olympians, binned); // rect mark
const plot = rect.plot({color: {legend: true}}); // plot!``````

TIP

If a transform isnâ€™t doing what you expect, try inspecting the options object returned by the transform. Does it contain the options you expect?

Transforms can derive channels (such as y above) as well as changing the default options. For example, the bin transform sets default insets for a one-pixel gap between adjacent rects.

Transforms are composable: you can pass options through more than one transform before passing it to a mark. For example, above itâ€™s a bit difficult to compare the weight distribution by sex because there are fewer female than male athletes in the data. We can remove this effect using the normalize transform with the sum reducer.

Fork
js
``````Plot.plot({
y: {percent: true},
marks: [
Plot.rectY(
olympians,
Plot.normalizeY(
"sum", // normalize each series by the sum per series
Plot.binX(
{y2: "count"}, // disable implicit stack transform
{x: "weight", fill: "sex", mixBlendMode: "multiply"}
)
)
)
]
})``````

And, as you may have wondered above, many of Plotâ€™s marks provide implicit transforms: for example, the rectY mark applies an implicit stackY transform if you use the y option, and the dot mark applies an implicit sort transform to mitigate the effect of occlusion by drawing the smallest dots on top.

## Custom transforms â€‹

For greater control, you can also implement a custom transform function, allowing data, indexes, or channels to be derived prior to rendering. Custom transforms are rarely implemented directly; see the built-in transforms above. For example, below we implement the filter transform â€śby handâ€ť as a custom transform to show the unemployment rates only in Michigan metropolitan divisions.

Fork
js
``````Plot.plot({
y: {
grid: true,
label: "Unemployment (%)"
},
color: {
domain: [false, true],
range: ["#ccc", "red"]
},
marks: [
Plot.ruleY([0]),
Plot.line(bls, {
x: "date",
y: "unemployment",
z: "division",
transform: (data, facets) => ({
data,
facets: facets.map((facet) => {
return facet.filter((i) => {
return /, MI /.test(data[i].division);
});
})
})
})
]
})``````

The transform function is passed three arguments, data, facets, and options representing the markâ€™s data and facet indexes, and the plotâ€™s options; it must then return a {data, facets} object with the transformed data and facet indexes. The facets are represented as a nested array of arrays such as [[0, 1, 3, â€¦], [2, 5, 10, â€¦], â€¦]; each element in facets specifies the zero-based indexes of elements in data that are in a given facet (i.e., have a distinct value in the associated fx or fy dimension).

If the transform option is specified, it supersedes any basic transforms (i.e., the filter, sort and reverse options are ignored). However, the transform option is rarely used directly; instead one of Plotâ€™s built-in transforms are used, and these transforms automatically compose with the basic filter, sort and reverse transforms.

While transform functions often produce new data or facets, they may return the passed-in data and facets as-is, and often have a side-effect of constructing derived channels. For example, the count of elements in a groupX transform might be returned as a new y channel. In this case, the transform is typically expressed as an options transform: a function that takes a mark options object and returns a new, transformed options object, where the returned options object implements the transform option. Transform functions should not mutate the input data or facets. Likewise options transforms should not mutate the input options object.

When implementing a custom transform for generic usage, keep in mind that it needs to be compatible with Plotâ€™s faceting system, which partitions the original dataset into discrete subsets.

## Custom initializers ^0.5.0â€‹

Initializers are a special class of transform; whereas transforms operate in abstract data space, initializers operate in screen space such as pixel coordinates and colors. For example, initializers can modify a marksâ€™ positions to avoid occlusion. Initializers are invoked after the initial scales are constructed and can modify the channels or derive new channels; these in turn may (or may not, as desired) be passed to scales. Plotâ€™s hexbin and dodge transforms are initializers.

You can specify a custom initializer by specifying a function as the mark initializer option. This function is called after the scales have been computed, and receives as inputs the (possibly transformed) array of data, the facets index of elements of this array that belong to each facet, the input channels (as an object of named channels), the scales, and the dimensions. The mark itself is the this context. The initializer function must return an object with data, facets, and new channels. Any new channels are merged with existing channels, replacing channels of the same name.

If an initializer desires a channel that is not supported by the downstream mark, additional channels can be declared using the mark channels option.

## transform(options, transform) ^0.4.3â€‹

js
``````Plot.transform(options, (data, facets) => {
return {
data,
facets: facets.map((I) => I.filter(() => Math.random() > 0.5))
};
})``````

Given an options object that may specify some basic transforms (filter, sort, or reverse) or a custom transform function, composes those transforms if any with the given transform function, returning a new options object. If a custom transform function is present on the given options, any basic transforms are ignored. Any additional input options are passed through in the returned options object. This method facilitates applying the basic transforms prior to applying the given custom transform and is used internally by Plotâ€™s built-in transforms.

## initializer(options, initializer) ^0.5.0â€‹

This helper composes the initializer function with any other transforms present in the options, and returns a new options object. It is used internally by Plotâ€™s built-in initializer transforms.

## valueof(data, value, type) â€‹

js
``Plot.valueof(aapl, "Close")``

Given an iterable data and some value accessor, returns an array (a column) of the specified type with the corresponding value of each element of the data. The value accessor may be one of the following types:

• a string - corresponding to the field accessor (`(d) => d[value]`)
• an accessor function - called as type.from(data, value)
• a number, Date, or boolean â€” resulting in an array uniformly filled with the value
• an object with a transform method â€” called as value.transform(data)
• an array of values - returning the same
• null or undefined - returning the same

If type is specified, it must be Array or a similar class that implements the Array.from interface such as a typed array. When type is Array or a typed array class, the return value of valueof will be an instance of the same (or null or undefined). When type is a typed array, values will be implicitly coerced numbers, and if type is Float64Array, Float32Array, or a subclass of the same, null values will be implicitly replaced with NaN. If type is not specified, valueof may return either an array or a typed array (or null or undefined).

valueof is not guaranteed to return a new array. When a transform method is used, or when the given value is an array that is compatible with the requested type, the array may be returned as-is without making a copy.

## column(source) ^0.4.3â€‹

js
``const [X, setX] = Plot.column();``

This helper for constructing derived columns returns a [column, setColumn] array. The column object implements column.transform, returning whatever value was most recently passed to setColumn. If setColumn is not called, then column.transform returns undefined. If a source is specified, then column.label exposes the given sourceâ€™s label, if any: if source is a string as when representing a named field of data, then column.label is source; otherwise column.label propagates source.label. This allows derived columns to propagate a human-readable axis or legend label.

This method is used by Plotâ€™s transforms to derive channels; the associated columns are populated (derived) when the transform option function is invoked.

## identity ^0.6.2â€‹

js
``Plot.contour(data, {width: w, height: h, fill: Plot.identity})``

This channel helper returns a source array as-is, avoiding an extra copy when defining a channel as being equal to the data.

## indexOf ^0.6.6â€‹

js
``Plot.lineY(numbers, {x: Plot.indexOf, y: Plot.identity})``

This channel helper returns an array of numbers [0, 1, 2, 3, â€¦]. It is used internally by marks with zero-based index defaults for channels.

Resources
Observable
Â© 2024 Observable, Inc. ISC License