Building Charts and Dashboards

The workspace

The Dashboard has a field list (the dataset's columns, split into dimensions and measures), a chart type palette, shelves where you drop fields, and a canvas of chart widgets. A dashboard can have several pages: use Add dashboard to create one, and double-click a tab to rename it, or use the tab's controls to duplicate or delete it. Duplicated pages contain independent widgets.

Choose the dataset a chart reads from the dataset selector. A calculated field belongs to the dataset it was written for.

Building a chart

1. Drag a chart type from the palette onto the canvas, or drop a field and let the Dashboard pick one.
2. Drag fields onto the shelves described below.
3. Click a measure to choose its aggregation: sum, average, count, min, max or median.
4. Move and resize the widget on the canvas. Each widget has edit and delete buttons.

The shelves

• Columns and Rows: the axes (category on one, measure on the other)
• Color: a dimension that splits each category into its own series, with a legend
• Size: a measure that drives bubble or point size (scatter, packed bubble)
• Shape: a dimension that drives marker shape (scatter)
• Detail: dimensions that add granularity or labels without visual encoding
• Tooltip: extra fields that appear only in the tooltip
• Play axis (scatter): a dimension, such as year, whose values become animation frames with a Play button and scrubber

Chart types

• Tables: Table, Matrix, Highlight Table
• Comparison: Column, Bar, Line, Clustered Column, Clustered Bar, Dual Line (two measures on independent axes), Combo (a line over columns), Bullet
• Distribution and relationship: Scatter (with optional bubble size), Histogram, Box Plot, Violin Plot, Packed Bubble, Word Cloud, Heatmap
• Composition: Donut, Stacked Column, 100% Stacked Bar and Column, Area, Treemap, Funnel, Waterfall, Decomposition Tree, Ribbon
• Flow and time: Network Diagram, Gantt
• Maps: Symbol Map, Choropleth (World countries or US states), Density Map, Azure Map
• Single values: Card, Multi-row Card, KPI, Gauge, Smart Narrative (a written summary of the data)
• Controls: Slicer and Button

Charts are drawn as sandboxed D3 visualisations, so a chart cannot reach the rest of the page.

Marks and formatting

Each chart has a settings area with sections including General, Color and Style, Axis and Legend, Tooltip, Labels, Actions and Filters, and Analytics.

• Sorting: by value (high to low, low to high), by the axis itself (A to Z, 1 to 9, and reversed), or in data order.
• Labels and totals: show values, show the second measure in the label, show totals, show or hide the legend.
• Number format: a plain number or currency, with a display unit of auto (K, M, B), none, thousands, millions or billions.
• Axes: titles, minimum and maximum, logarithmic scale, and label colours.
• Colour: a palette, a gradient scale, or fixed thresholds, plus Conditional formatting. A Color Vision Preview shows the chart as people with deuteranomaly, protanomaly, tritanomaly or no colour vision would see it, so you can check it stays readable.
• Backgrounds: a colour or an image, and a title colour.

Filters, slicers and interactions

• Filters have three scopes: this visual, this page and all pages. Drop a field in the zone you want, then click its chip to choose the operator and values. To change scope, remove the chip and drop it in another zone.
• Slicers are controls (shown as a list or as tiles) that filter the other charts on the page.
• Cross-filtering: click a bar, slice or point to filter the other charts on the page; click the same one again to clear the filter.
• Drill-down: with more than one dimension on Columns, clicking a mark descends to the next level of the hierarchy; the widget remembers the path so you can step back up.
• Drill-through: right-click a page tab to make it a drill-through target and list the field names it accepts (for example region, product). Then right-click a mark on another page and choose the target: the target page opens filtered to that value, with a Back pill to return.
• URL actions open a link built from a field, such as https://example.com/{field}.
• Bookmarks save the current filter, cross-filter and slicer state. Stories replay them; see Calculations, parameters and stories.

Data versions and refreshing

A line under the toolbar says which version of each dataset the dashboard draws from. By default it follows the newest. Pin a version to hold the charts steady, or follow again to catch up. A pipeline can refresh the dashboard with new data using Refresh Dashboard, which checks that the dashboard's fields still exist first. If you edited the layout since the last refresh, a banner says so.

Exporting

Export data and chart settings downloads the chart's data and its settings as JSON, so a figure can be reproduced or inspected. Export PNG captures the whole dashboard as dashboard.png, using the current light or dark theme. Wait for the charts to finish rendering first, or the export reports that a chart did not respond. Research figures include the provenance of the run that produced them.

Training metrics and the Research module

Loss, accuracy and resource usage from a training run appear on the Dashboard while it runs. The Research module also publishes measured figures here as ordinary editable widgets, with frozen provenance for the run behind each one.

Tips

• Keep one dataset per page where you can; it makes filters predictable.
• Put the dimension on Columns and measures on Rows for most comparison charts.
• Use Detail instead of Color when you want more points but a single colour.