Filters

kepler's filter panel is unchanged in the panel. What is worth thinking about in a dashboard is which of the two available filters to use — because there are now two, and they behave very differently.

Two places to narrow data

	Query filter (WHERE)	kepler filter
Runs	in the database	in the browser
Rows removed	never leave the database	are still loaded, just hidden
Cost of changing it	a query	nothing
Can be widened again	yes, re-queries	yes, instantly
Affects other panels	if it uses a shared variable	only through cross-filtering

The rule of thumb: use the query to decide what data comes to the browser, and kepler filters to explore it once it is here.

A kepler filter is a mask. The rows are still loaded, so dragging a slider is free and reversible. That is exactly what you want for exploration, and exactly what you do not want for reducing a hundred million rows to a manageable set.

The time filter

A dataset with a time column gets a time filter with a histogram and a brush. It is the same mechanism as any other filter, with one difference: it carries a fixed domain, so narrowing the brush does not rescale the histogram underneath it.

That property is what makes the Slider writes variables time-sync mode work — the brush stays a mask over the whole dataset, so you can always widen it again. See Time range sync.

The time widget's settings choose a 12- or 24-hour clock (Time Format) beside the time zone.

A numeric filter as a timeline

A range filter's card has a clock button, Time Playback, that opens the filter as a timeline at the bottom of the map, like the time filter's: a histogram of the column in bins (Count of Rows by mag (0.1 bins) on a map of earthquakes), a brush, play, speed and the playback window menu. The widget's settings set the bin size, and a Y axis can plot a column's average or another aggregate instead of the count. Playing it sweeps the window across magnitudes, depths or prices the way the time filter sweeps across time.

Only one filter is a timeline at a time: opening one sends any other, the time filter included, back to the side panel. It still filters there, and time range sync still moves it; only its bar is gone. The panel's time features (time range sync, peer time sync, the time window variables) read the time filter only, never a numeric timeline. A range Filter mapping on the same column under Cross-filtering reads its window as it reads any range filter's. See Time playback.

Filter out

A filter's Filter out switch turns it inside out: the rows that match are hidden and the rest stay. Select the daytime hours and switch it on to keep the night; pick a category to see everything but it. It works on range, select, time and polygon filters.

One cost to know on a large dataset: kepler runs an inverted range or time filter on the CPU rather than the GPU, so dragging one is slower than dragging the same filter the right way out. A numeric timeline set to Step by Interval runs on the CPU too.

One filter across datasets

A filter can act on several datasets at once, matching a field in each — Add synced filter in the filter's card. Time filters always could; range and select filters can too now, so a map built from two queries that share a column, a site or a category, narrows both with one control.

Publishing a filter to the dashboard

A select, multi-select or text filter can be bound to a dashboard variable, and a numeric range filter to a pair of them, so filtering the map filters every other panel. This is the two-way channel: changing the variable also applies the filter on the map.

Set it up under Cross-filtering in the panel options, not in kepler's panel — see Cross-filtering.

Polygon filters and the draw tool

A shape drawn with the draw tool can be applied as a polygon filter on a layer. It can also be published to a dashboard variable as WKT, which is a separate thing — the publishing happens as soon as the figure is finished, whether or not you go on to make it a filter. See Publishing the viewport and drawn areas.

Filters are part of the saved configuration

Save current map captures the filters along with everything else, so a dashboard can open with a filter already applied. Worth remembering when a panel appears to be missing data that the query clearly returns — check the filter panel before checking the query.

---

Upstream: Filters

Written against kepler.gl 3.3.0-alpha.20. See Upstream documentation.