navio

Navio:
A visualization widget to understand and explore your data

Use it to summarize, explore and navigate your multivariate data using three simple interactions:

Sort	Filter a Range	Filter By Value
Click on a header to sort	Drag to select a range	Click on a value to select all instances

Try it!

You can test Navio right now with your own CSV or JSON data (less than 200MB), using:

Obervable Notebook	Shipyard	Jupyter Notebook

Other demos:

MoMa Collection (Code)
Navio-only Vast 2017 MiniChallenge1 (Code)
Co-voting patterns of the Colombian senate
Simplest example with Networks on SVG (Code)
Simple example with Networks on Canvas (Code)
IEEEVIS Publications Data (Code) (Observable Notebook)
Horizontal and vertical, side by side (Code)
Inside an Observable notebook (Code) — runs through the real @observablehq/runtime, in the cell layout a notebook actually produces
Bound to another widget with Inputs.bind (Code)

Comparing

Why using something else for summarizing your data?. Here is how Navio compares with other alternatives:

Navio vs Parallel Coordinates

You can use this Notebook to compare Navio with Parallel Coordinates, using your own data. Please be aware that the Vegalite implementation of Parallel Coordinates will break with a few thousand rows (on the image below it broke with 500 rows and 86 attributes of the fifa19 Kaggle Dataset)

Navio vs Scatterplot Matrix

Use this Notebook to compare Navio with a Scatterplot Matrix, using your own data. Please be aware that the Vegalite implementation of the Scatterplot Matrix only support quantitative attributes and will also break with a dozen attributes and a few hundred rows), therefore the image below only displayed 8 attributes (out of the 28) on the scatterplot matrix.

Install

npm install navio

Or use it from unpkg

  <script type="text/javascript" src="https://d3js.org/d3.v6.min.js"></script>
  
  <script src="https://unpkg.com/popper.js@1.14/dist/umd/popper.min.js"></script>
  <script type="text/javascript" src="https://unpkg.com/navio/dist/navio.min.js"></script>

Requires [^popper.js@0.14](https://github.com/FezVrasta/popper.js/), [^d3@4.13](http://d3js.org). If you want to use d3@4 use navio@0.0.67

Usage

TLDR

<!DOCTYPE html>
<body>
  <!-- Placeholder for the widget -->
  <div id="navio"></div>

  <!-- NAVIO Step 0: Load the libraries -->
  <script type="text/javascript" src="https://d3js.org/d3.v6.min.js"></script>
  <script src="https://unpkg.com/popper.js@1.14/dist/umd/popper.min.js"></script>
  <script type="text/javascript" src="https://unpkg.com/navio/dist/navio.min.js"></script>

<script>
  // NAVIO  Step 1.  Create a Navio passing a d3 selection to place it and an optional height
  var nv = navio(d3.select("#navio"), 600);

  d3.csv(YOUR_DATA).then(data) => {
    // NAVIO Step 2. Load your data!
    nv.data(data);

    // NAVIO Step 3. Detect your attributes (or load them manually)
    nv.addAllAttribs();

    // Optional, setup a selection callback
    nv.updateCallback( selected => console.log("selected in Navio: ", selected.length));
  });
</script>
</body>
</html>

Step by step

HTML. Start with this template ```html <!DOCTYPE html>

Basic Usage // Your Navio widget goes here

2. **Import Navio**. Create and import a new JavaScript file below the scripts (d3 and Navio) or right in the html like in the example below.
```html
<script src="https://d3js.org/d3.v6.min.js"></script>
<script type="text/javascript" src="https://unpkg.com/navio/dist/navio.min.js"></script>
<script type="text/javascript">
  //   YOUR_JS_CODE_HERE
</script>

Create a Navio Instance

var nv = navio(d3.select("#Navio"), 600); //height 600

[Optional] Configure navio to your liking

// Default parameters
nv.x0 = 0;  //Where to start drawing navio in x
nv.y0 = 100; //Where to start drawing navio in y, useful if your attrib names are too long
nv.maxNumDistinctForCategorical = 10; // addAllAttribs uses this for deciding if an attribute is categorical (has less than nv.maxNumDistinctForCategorical categories) or ordered
nv.maxNumDistinctForOrdered = 90; // addAllAttribs uses this for deciding if an attribute is ordered (has less than nv.maxNumDistinctForCategorical categories) or text. Use nv.maxNumDistinctForOrdered = Infinity for never choosing Text

nv.howManyItemsShouldSearchForNotNull = 100; // How many rows should addAllAttribs search to decide guess an attribute type
nv.margin = 10; // Margin around navio

nv.levelsSeparation = 40; // Separation between the levels
nv.divisionsColor = "white"; // Border color for the divisions
nv.levelConnectionsColor = "rgba(205, 220, 163, 0.5)"; // Color for the conections between levels
nv.divisionsThreshold = 4; // What's the minimum row height needed to draw divisions
nv.fmtCounts = d3.format(",.0d"); // Format used to display the counts on the bottom
nv.nestedFilters = true; // Should navio use nested levels?

nv.showAttribTitles = true; // Show headers?
nv.attribWidth = 15; // Width of the columns
nv.attribRotation = -45; // Headers rotation
nv.attribFontSize = 13; // Headers font size
nv.attribFontSizeSelected = 32; // Headers font size when mouse over

nv.filterFontSize = 10; // Font size of the filters explanations on the bottom

nv.tooltipFontSize = 12; // Font size for the tooltip
nv.tooltipBgColor = "#b2ddf1"; // Font color for tooltip background
nv.tooltipMargin = 50; // How much to separate the tooltip from the cursor
nv.tooltipArrowSize = 10; // How big is the arrow on the tooltip

nv.digitsForText = 2; // How many digits to use for text attributes

nv.id("attribName"); // Shows this id on the tooltip, should be unique

nv.addAllAttribsRecursionLevel = Infinity; // How many levels depth do we keep on adding nested attributes
nv.addAllAttribsIncludeObjects = false; // Should addAllAttribs include objects
nv.addAllAttribsIncludeArrays = false; // Should addAllAttribs include arrays

nv.DEBUG = false; // Set to true to trace Navio's internals to the console. Navio is
// silent by default; genuine problems (bad arguments, skipped
// attributes) are always reported via console.warn regardless.

// Default colors for values
nv.nullColor = "#ffedfd"; // Color for null values
nv.defaultColorInterpolator = d3.interpolateBlues;
nv.defaultColorInterpolatorDate = d3.interpolatePurples;
nv.defaultColorInterpolatorDiverging = d3.interpolateBrBG;
nv.defaultColorInterpolatorOrdered = d3.interpolateOranges;
nv.defaultColorInterpolatorText = d3.interpolateGreys;
nv.defaultColorRangeBoolean = ["#a1d76a", "#e9a3c9", "white"]; //true false null
nv.defaultColorRangeSelected = ["white", "#b5cf6b"];
nv.defaultColorCategorical = d3.schemeCategory10;

// // Discouraged: If you want to break perceptual rules to have many more categories use
// // the following "Piñata mode 🎉"
// nv.defaultColorCategorical = d3.schemeCategory10
//   .concat(d3.schemeAccent)
//   .concat(d3.schemePastel1)
//   .concat(d3.schemeSet2)
//   .concat(d3.schemeSet3);
// nv.maxNumDistinctForCategorical = nv.defaultColorCategorical.length;

[Optional] Add your attributes manually. Navio supports six types of attributes: categorical, sequential (numerical), diverging (numerical with negative values), text, date and boolean. You can either add them manually or use nv.addAllAttribs() to auto detect them (must be called after seting the data with nv.data(your_data))

nv.addCategoricalAttrib("attribName", [customScale]);
nv.addSequentialAttrib("attribName", [customScale]);
nv.addDivergingAttrib("attribName", [customScale]);
nv.addTextAttrib("attribName", [customScale]); // Colors by the first nv.digitsForText
nv.addOrderedAttrib("attribName", [customScale]); // Sorts and then colors by rank
nv.addDateAttrib("attribName", [customScale]);
nv.addBooleanAttrib("attribName", [customScale]);

If you ommit the [customScale] parameter it will use the defaults. You can also create your own custom made parameters using nv.addAttrib("attribName", customScale). For example, if you already have a scale for setting the colors of a cluster property on your visualization, you can tell navio to use the same matching colors. Make sure to set the domain and range of the scale, as navio will not try to do it with this function.

var color = d3.scaleOrdinal(d3.schemeSet3)
  .domain["cluster1", "cluster2", "cluster3"];

nv.addAttrib("cluster", color);

Set the data

After loading your data pass it to navio. This will trigger the drawing operation. You can force redrawing using nv.update();

nv.data(myData);

If your data is a network, or you have some links in the same format of a d3.forceSimulation you can also add them to navio using nv.links([links]). This won’t trigger a redraw, so make sure to call it before setting your data

nv.links(myLinks);
nv.data(myData);

Detect Attributes. navio also includes a function that detects the attributes automatically, which is slow, redraws the whole thing, and my be buggy. Use it at your own risk. But make sure to call it after setting your data

nv.data(myData);
nv.addAllAttribs();

Set a callback. A function that navio will call when the user filters/sort the data
```
nv.updateCallback( data => console.log("The filtered data is ", data));
```

Other methods

# nv.update() <>

Use it to force a redraw of navio after changing the underlying data without losing the filters. Useful in case you modify the data with some other action in your code, e.g. you recomputed clusters in a network chart.

# nv.hardUpdate([opts]) <>

Slower update that recomputes brushes and checks for parameters. Use it if you change any parameters or added new attributes after calling .data. opts can be an object that contains any of the following attributes:

shouldDrawBrushes (defaults true)
shouldUpdateColorDomains (defaults true)
recomputeBrushes (defaults true)
levelsToUpdate (defaults all levels, should be an array of indices)

# nv.getColorScale(attr ) <>

Returns the color scale for a certain attribute, make sure to pass an attribute that has been already added

# nv.getAttribs( ) <>

Returns the ordered list of attributes added to navio

# nv.isSelected(rowOrIndex)

Whether a row is currently selected. Accepts either one of the row objects you passed to .data() or its index into that array.

# nv.getRowsAtLevel([level = 0])

The rows present at a level, in the order they are drawn. Use this to observe the visual ordering produced by sorting.

Options

Every option can be set at construction, which also gets the ordering right for the ones that are only read once:

const nv = new navio(d3.select("#navio"), {
  height: 600,
  attribWidth: 20,
  orientation: "vertical",
  tooltipBgColor: "#eee",
});

A bare number is still the height, so the original new navio(selection, 600) is unchanged. The same object works for NavioWidget(data, options).

nv.getOptions()        // every option and its value; round-trips back in
nv.setOptions({ ... }) // apply to a live instance and redraw

An unrecognised key warns rather than being silently ignored — attribWidht: 20 used to land as a dead property and do nothing. So does naming a column that is not in your data (nv.addCategoricalAttrib("speceis") used to draw a stripe of nulls) or sorting by one that was never added.

Renamed: maxNumDistictForCategorical and maxNumDistictForOrdered were misspelled; they are now maxNumDistinct.... The old names still work and warn. legendFont has been removed — it had no effect on anything.

Which build am I running?

Navio prints its version once per page load, and exposes it:

navio.version; // e.g. "0.1.6"

Worth checking when loading from a CDN — https://unpkg.com/navio/dist/navio.min.js follows whatever is latest, and notebooks cache aggressively.

Debugging

nv.DEBUG = true traces Navio’s internals to the console. It is per-instance and only reachable after construction, so it misses everything logged while the widget is being built. To trace from the very first call, set the default before constructing:

navio.DEBUG = true;
const nv = new navio(el, 400); // traces construction and the first data()

or set a global before the script loads at all — useful from a devtools console plus a reload, or an Observable cell that runs ahead of the require:

window.NAVIO_DEBUG = true;

Either way the default stays off, so no rebuild is needed to turn tracing on.

Settings panel

A gear in the widget’s corner opens a panel for changing options live: which columns are shown, their order, orientation, and the geometry.

nv.settings = false;             // hide the gear entirely (default true)
nv.settingsPlacement = "beside"; // "below" (default) | "beside" | "over"
nv.settingsMaxAttribRows = 10;   // scroll the column list past this many; 0 disables

The panel is a real <dialog>, and every placement keeps the widget visible so you can watch the effect of each control as you change it. "below" is the default because column width changes the canvas width, so a panel below does not move while you drag that slider; "beside" sits to the right of the canvas; "over" is a compact overlay for layouts with no room. You can also switch placement from inside the panel itself.

A panel will not reposition while you are holding one of its own controls — the height slider makes the widget taller, and without that the slider would walk down the page away from the cursor dragging it.

The Attributes section is a <details>, so a table with many columns can be folded away to reach Layout, Colours and Filtering; past settingsMaxAttribRows columns the list scrolls inside its own box, with Show all / Show none left outside it. Every parameter carries a description on hover.

Reset puts the widget back the way it was constructed and forgets what was saved — nv.resetSettings(). Filters and the selection are deliberately left alone, so a layout can be reset without discarding a selection you have already made; use setFilters() for those.

Hiding a column is not removing it. It keeps its type, its colour scale and its place in the order — and any selection you have already made survives, because filters are materialised when created and hiding touches only the layout. Sorting is in place, so hiding the sorted column leaves the order alone too.

The panel also changes an attribute’s type — how a column is interpreted and coloured. addAllAttribs guesses from the data and sometimes guesses wrong; this is the correction. Only the colouring changes: the column keeps its position, and sorting and filtering compare raw values, so nothing you have already selected is invalidated.

nv.getAttribType("beak");             // "seq"
nv.setAttribType("beak", "ordered");  // cat seq ordered text date div bool
nv.getAttribTypes();                  // the switchable set, for a picker

Reorder by dragging an attribute name in the panel, or with the ↑ ↓ buttons (which is the keyboard path). Dragging a column header on the widget itself works too, and the panel follows it.

Sorting vs reordering: a click on a header sorts; Shift-drag its label reorders. They are on different modifiers rather than told apart after the fact, because a rotated label makes a plain click hard to distinguish from a small drag — the two used to interfere and clicks would silently do nothing.

nv.setAttribVisible("beak", false);  // hide one
nv.getVisibleAttribs();              // what is drawn, in order
nv.getHiddenAttribs();               // names currently hidden
nv.setHiddenAttribs([]);             // show everything again

None of these emit a change event, so a widget bound with Inputs.bind is unaffected.

Panel changes are remembered in localStorage and reapplied on the next load. Direct property assignments cannot notify anyone, so call nv.saveSettings() after those if you want them remembered.

The key is scoped to the page — navio.settings.<origin><pathname>.<slot> — so a widget in one notebook no longer inherits the column layout of a widget in an unrelated one. Within a page, instances are told apart by the container’s own id when it has one, falling back to construction order. The query string and hash are deliberately left out, so filtering the page or following a #anchor is still the same widget.

nv.settingsKey = "my-page";   // storage key; null turns persistence off
nv.getSettings() / nv.setSettings(cfg)
nv.getSettingsCode()          // the JS that reproduces the current settings
nv.clearStoredSettings()

The panel’s Copy config button puts getSettingsCode() on the clipboard, so a layout you arrived at by fiddling can be pasted straight into your source.

The picker is pluggable — nv.attribPicker takes any (names, {value, onChange, move}) => HTMLElement. See examples/settings, which plugs in @john-guerra/search-checkbox for search plus All/None. Navio itself takes no dependency on it.

Orientation

nv.orientation = "vertical"; // default "horizontal"
nv.hardUpdate();

Navio has two logical axes: the attribute axis, along which the columns are laid out, and the record axis, along which one line per row is drawn. horizontal (the default, and the historical behaviour) puts attributes across and records down. vertical transposes them — attributes become labelled rows and records run left to right, with drill-down levels stacking downward.

Everything works in both: sorting, click-to-filter, brushing a range (drag along whichever axis the records run), drill-down, and the tooltip.

Accessibility

Navio draws to a canvas, so its content is opaque to assistive technology. The controls around it are labelled and keyboard-operable:

Interaction	Mouse	Keyboard
Sort a column	click the header	focus the header, `Enter`
Reorder a column	drag the header	`Alt` + `←`/`→`
Remove a filter	click the chip	focus the chip, `Enter`
Close a level	click the ✕	focus it, `Enter`

Selection and filter changes are announced through an aria-live region, and header hover transitions are skipped when prefers-reduced-motion: reduce is set.

Brushing a range still has no keyboard equivalent — see #68.

A note on your data

Navio does not add any properties to the rows you give it. Earlier versions wrote selected, __i and __seqId onto every row; that bookkeeping now lives in typed arrays inside the instance, so your objects stay exactly as you passed them and two Navios can share one array safely.

If you were reading d.selected, call nv.isSelected(d) instead. If you were reading d.__i, use nv.getRowsAtLevel(level) to get the drawn order.

Navio can be used as a Reactive Widget: an HTML element that holds its state in .value and emits an input event whenever the user changes it. That makes it bindable to other widgets, and usable directly as an Observable viewof.

import { NavioWidget } from "navio";

const w = NavioWidget(data, { height: 600 });
document.body.appendChild(w);

w.addEventListener("input", () => render(w.value));

.value is the array of selected rows — the ones surviving every level of the drill-down. That is what makes it work as an Observable viewof:

viewof selected = NavioWidget(data, { height: 400 })
Inputs.table(selected)                            // the rows, as expected
Inputs.bind(Inputs.table(data), viewof selected)  // and it binds both ways

Assigning rows selects them. They are matched by nv.id(), so a peer holding the same data round-trips; with the default id they are matched by object identity, so call nv.id("someKey") to sync across instances.

w.value = data.filter((d) => d.mass > 4000); // selects them, emits nothing

The filter chain

The rows are the output; what Navio actually manipulates is a multi-level filter chain, one entry per level:

[
  [{ type: "value", attrib: "species", value: "Adelie" }],
  [{ type: "value", attrib: "island",  value: "Torgersen" }],
]

This, not the row list, is the form that replays faithfully onto another instance: rows are projections through this instance’s arrays, whereas the chain describes how they were chosen. It is also JSON-safe, so it can go in a URL or localStorage.

w.getFilters();                       // the chain
w.setFilters(JSON.parse(saved));      // restore it; assigning one works too
w.value = { filters: chain };         //   ...either form
w.snapshot();                         // { filters, selection } in one read

Assigning a chain (or a { filters } wrapper) applies it and .value settles back on the rows it produced. Anything else is read as rows.

The classic navio(selection, height) API is unchanged; this is additive.

# nv.destroy()

Tears the instance down: removes its tooltip, detaches the listeners it added to document.body, empties its container, and drops its reference to your data.

Call this whenever you unmount a Navio in a single-page app (React, Vue, Svelte, Observable). Without it, the listeners on body keep the whole instance — including the dataset — reachable, so the memory is never reclaimed. Other Navio instances on the page are unaffected, and calling it twice is safe.

// React
useEffect(() => {
  const nv = new navio(d3.select(ref.current), 600);
  nv.data(myData);
  return () => nv.destroy();
}, []);

License

Navio.js is licensed under the MIT license. (http://opensource.org/licenses/MIT)

This site is open source. Improve this page.