library(plotly.core)
library(crosstalk)
library(data.table)
#>
#> Attaching package: 'data.table'
#> The following object is masked from 'package:base':
#>
#> %notin%
gap <- px_data_gapminder() # one row per country and year (1952-2007)crosstalk links htmlwidgets in the browser, with no Shiny server: selecting points in one figure selects the same rows in every other view of the data. This page shows how to do it with go_figure() and add_<trace>(); px_*() functions work the same way when given a SharedData as data_frame.
The examples use the gapminder data, with a row for each country every five years. Rows are keyed by country, so selecting any point of a country selects all of its years, in every view. Views of the same SharedData share one selection, so each example makes its own: selections in one example don’t reach the others.
Linking traces
Wrap the data in a SharedData and pipe it into go_figure(). Formula properties (x = ~gdpPercap) are evaluated in its data, as with a data frame, and each trace built from its rows is linked: every point carries its row’s key.
SharedData$new(gap, key = ~country) |>
go_figure() |>
add_scatter(x = ~gdpPercap, y = ~lifeExp, text = ~country, mode = "markers") |>
update_layout(xaxis_type = "log") |>
highlight(color = "red")Click a point to select its country: its points from all twelve years are highlighted, showing its path. Double-click to clear; hold shift to add more countries.
-
keysays which rows belong together; the default is the data’s row names (one key per row). - Views sharing a
group(by default, one perSharedData; set it withSharedData$new(group = )) are linked. - Without
highlight(), a linked figure selects on click and clears on double-click.
Several views of the same rows
Each figure made from the same SharedData, or from SharedDatas with the same group and keys, is a view of the same selection. Here 2007’s rows and all years’ rows are linked through the group “countries”: selecting countries with a box or lasso in the first plot shows their histories in the second, where attrs_selected() draws the selected points as lines.
in_2007 <- SharedData$new(gap[year == 2007], key = ~country, group = "countries")
all_years <- SharedData$new(gap, key = ~country, group = "countries")
latest <- in_2007 |>
go_figure() |>
add_scatter(x = ~gdpPercap, y = ~lifeExp, text = ~country, mode = "markers") |>
update_layout(height = 350, xaxis_type = "log", title = "Wealth and life expectancy, 2007") |>
highlight(on = "plotly_selected", color = "red")
history <- all_years |>
go_figure() |>
add_scatter(x = ~year, y = ~lifeExp, text = ~country, mode = "markers") |>
update_layout(height = 350, title = "Life expectancy over time") |>
highlight(color = "red", selected = attrs_selected(mode = "lines+markers"))
htmltools::tagList(as_widget(latest), as_widget(history))Clicking a point of the history plot selects that country in both.
Which traces are linked
An add_<trace>() call links its trace when one of the trace’s data arrays (x, y, z, values, ids, …) has one value per row of the SharedData. Other traces, such as summaries and reference lines, stay unlinked and unchanged by selections:
world <- gap[, .(lifeExp = weighted.mean(lifeExp, pop)), by = year]
#> Warning in gsum(w, na.rm = FALSE): The sum of an integer column for a group was
#> more than type 'integer' can hold so the result has been coerced to 'numeric'
#> automatically for convenience.
SharedData$new(gap, key = ~country) |>
go_figure() |>
add_scatter(x = ~year, y = ~lifeExp, text = ~country, mode = "markers", name = "countries") |>
add_scatter(x = world$year, y = world$lifeExp, mode = "lines", name = "world") |>
highlight(color = "red", selected = attrs_selected(mode = "lines+markers"))A SharedData can also be given to a single trace with data:
go_figure() |>
add_scatter(x = ~year, y = ~lifeExp, data = all_years, mode = "markers") |> # linked
add_scatter(x = ~year, y = ~pop, data = gap, mode = "markers") # not linkedLinking traces built from other data
set_crosstalk_keys() links traces whose values don’t come from the SharedData’s rows, such as summaries: give it the keys of the trace’s points and the group to join. Here the points are keyed by continent, and a bar chart of countries per continent is linked to them, so clicking a bar selects every country of that continent (and clicking a point selects its continent):
by_continent <- SharedData$new(gap, key = ~continent)
countries <- gap[year == 2007, .N, by = continent]
continent_bars <- go_figure() |>
add_bar(x = countries$continent, y = countries$N) |>
set_crosstalk_keys(countries$continent, group = by_continent$groupName()) |>
update_layout(height = 300, yaxis_title = "countries") |>
highlight(color = "red")
continent_points <- by_continent |>
go_figure() |>
add_scatter(x = ~gdpPercap, y = ~lifeExp, text = ~country, mode = "markers") |>
update_layout(height = 350, xaxis_type = "log") |>
highlight(color = "red")
htmltools::tagList(as_widget(continent_bars), as_widget(continent_points))With simple = TRUE, a trace stands for all its keys as a whole, for marks that can’t be split into points, such as a line summarising a group. Here each continent’s average life expectancy is a line standing for its countries: clicking a line selects them all in the plot below it. A line is highlighted when all of its countries are selected.
by_country2 <- SharedData$new(gap, key = ~country)
averages <- go_figure()
for (cont in sort(unique(gap$continent))) {
d <- gap[continent == cont, .(lifeExp = mean(lifeExp)), by = year]
averages <- averages |>
add_scatter(x = d$year, y = d$lifeExp, mode = "lines+markers", name = cont) |>
set_crosstalk_keys(unique(gap[continent == cont, country]),
group = by_country2$groupName(),
traces = length(averages$data) + 1L, simple = TRUE)
}
# plotly.js reports clicks on a line's points: let a click near the line count
averages <- averages |>
update_layout(height = 350, hoverdistance = 50, title = "Average by continent") |>
highlight(color = "red")
country_points <- by_country2 |>
go_figure() |>
add_scatter(x = ~year, y = ~lifeExp, text = ~country, mode = "markers") |>
update_layout(height = 300, title = "Countries") |>
highlight(color = "red")
htmltools::tagList(as_widget(averages), as_widget(country_points))Highlighting options
highlight() sets how a figure makes and shows selections:
| Argument | Effect |
|---|---|
on |
Event that selects: "plotly_click" (default), "plotly_hover", "plotly_selected" (box/lasso; sets the drag mode to select), "plotly_selecting" (while dragging) |
off |
Event that clears: matches on by default ("plotly_deselect" for selections, else "plotly_doubleclick") |
color |
Selection colour (R colour, hex or CSS such as "rgba(255,0,0,0.5)"); NULL keeps the points’ colours |
dynamic |
Adds a colour picker, for selections in several colours |
persistent |
New selections add to the current one (as holding shift does) |
selectize |
Adds a searchable dropdown of the keys; or a list of selectize.js options |
defaultValues |
Keys selected when the page opens |
opacityDim |
Opacity multiplier of unselected points (default 0.2) |
selected |
Attributes of selected points, e.g. attrs_selected(mode = "lines+markers")
|
debounce |
Milliseconds to wait, for frequent events such as hover |
Compare countries: pick them in the dropdown (or click their points), change the colour between picks, and they are kept (persistent = TRUE):
SharedData$new(gap, key = ~country) |>
go_figure() |>
add_scatter(x = ~year, y = ~lifeExp, text = ~country, mode = "markers") |>
highlight(
dynamic = TRUE,
persistent = TRUE,
selectize = TRUE,
defaultValues = c("China", "India"),
selected = attrs_selected(mode = "lines+markers")
)Filters
crosstalk’s filter inputs subset every view of a SharedData’s group: traces show only the rows that pass all filters. Filters work on keys (a key passes when any of its rows does), so filter by year with a key per row, the default:
filtered <- SharedData$new(gap)
htmltools::tagList(
filter_checkbox("continent", "Continent", filtered, ~continent, inline = TRUE),
filter_slider("year", "Year", filtered, ~year, width = "100%", sep = ""),
filtered |>
go_figure() |>
add_scatter(x = ~gdpPercap, y = ~lifeExp, text = ~country, mode = "markers") |>
update_layout(height = 350, xaxis_type = "log") |>
as_widget()
)Notes
-
Subplots: traces added to several cells (
add_scatter(..., row = 1, col = "all")) are each linked.set_crosstalk_keys()uses trace indices, so call it after adding the traces it links. -
Animation: frames made by
px_*()functions are linked frame by frame. - Other widgets: any crosstalk-compatible widget, such as DT tables and leaflet maps, can share the group.
- Trace types: selection events need traces with points (scatter, bar, histogram, box, maps, …); plotly.js reports clicks and hovers on a line’s points, not between them. Heatmaps, pies and other aggregating traces can’t be selected point by point.
-
Shiny: crosstalk works without a server; in Shiny,
event_data()also reports selections, andSharedDataselections can be read with<SharedData>$selection(). - Size: keys are sent as strings, one per point; with many thousands of points, short keys keep pages small.