Inspecting Data

Makie provides a data inspection tool via DataInspector(x) where x can be a figure, axis or scene. With it you get a floating tooltip with relevant information for various plots by hovering over one of its elements.

By default the inspector will be able to pick any plot other than text and volume based plots. If you wish to ignore a plot, you can set its attribute plot.inspectable[] = false. With that the next closest plot (in range) will be picked.

DataInspector(figure_axis_or_scene = current_figure(); kwargs...)

Creates a data inspector which will show relevant information in a tooltip when you hover over a plot.

This functionality can be disabled on a per-plot basis by setting plot.inspectable[] = false. The displayed text can be adjusted by setting plot.inspector_label to a function (plot, index, position) -> "my_label" returning a label. See Makie documentation for more detail.

Keyword Arguments:

  • range = 10: Controls the snapping range for selecting an element of a plot.

  • priority = 100: The priority of creating a tooltip on a mouse movement or scrolling event.

  • enabled = true: Disables inspection of plots when set to false. Can also be adjusted with enable!(inspector) and disable!(inspector).

  • indicator_color = :red: Color of the selection indicator.

  • indicator_linewidth = 2: Linewidth of the selection indicator.

  • indicator_linestyle = nothing: Linestyle of the selection indicator

  • enable_indicators = true): Enables or disables indicators

  • depth = 9e3: Depth value of the tooltip. This should be high so that the tooltip is always in front.

  • apply_tooltip_offset = true: Enables or disables offsetting tooltips based on, for example, markersize.

  • and all attributes from Tooltip

Custom text

The text that DataInspector displays can be adjusted on a per-plot basis through the inspector_label attribute. It should hold a function (plot, index, position) -> "my_string", where plot is the plot whose label is getting adjusted, index is the index returned by pick (see events documentation) and position is the position of the inspected object.

lbls = ["Type A", "Type B"]
fig, ax, p = scatter(
    rand(10), color = rand(1:2, 10), colormap = [:red, :blue],
    inspector_label = (self, i, p) -> lbls[self.color[][i]]
)
DataInspector(fig)
fig

Extending DataInspector

The inspector implements tooltips for primitive plots and a few non-primitive plots (i.e. recipes). All other plots fall back to tooltips of one of their child plots.

For example a poly consists of a mesh and a wireframe plot, where wireframe is implemented as lines. Since neither poly nor wireframe has a specialized show_data method, DataInspector uses either mesh or lines to generate the tooltip.

While this means that most plots have a tooltip it also means many may not have a fitting one. If you wish to implement a more fitting tooltip for a new plot type you can do so by extending

function show_data(inspector::DataInspector, my_plot::MyPlot, idx, primitive_child::SomePrimitive)
    ...
end

Here my_plot is the plot you want to create a custom tooltip for, primitive_child is one of the primitives your plot is made from (scatter, text, lines, linesegments, mesh, surface, volume, image or heatmap) and idx is the index into that primitive plot. The latter two are the result from pick_sorted at the current mouseposition. In general you will need to adjust idx to be useful for MyPlot.

Let's take a look at the BarPlot method, which also powers hist. It contains two primitive plots - Mesh and Lines. The idx from picking a Mesh is based on vertices, of which there are four per rectangle. From Lines we get an index based on the end point of a line. To draw the outline of a rectangle as is done in barplot, we need 5 points and a separator totaling 6. We thus implement

import Makie: show_data

function show_data(inspector::DataInspector, plot::BarPlot, idx, ::Lines)
    return show_barplot(inspector, plot, div(idx-1, 6)+1)
end

function show_data(inspector::DataInspector, plot::BarPlot, idx, ::Mesh)
    return show_barplot(inspector, plot, div(idx-1, 4)+1)
end

to map the primitive idx to one identifying the bars in BarPlot. With this we can now get the position of the hovered bar with plot[1][][idx]. To align the tooltip to the selection we need to compute the relevant position in screen space and update the tooltip position.

using Makie: parent_scene, shift_project, update_tooltip_alignment!, position2string

function show_barplot(inspector::DataInspector, plot::BarPlot, idx)
    # Get the tooltip plot
    tt = inspector.plot

    # Get the scene BarPlot lives in
    scene = parent_scene(plot)

    # Get the hovered data-space position
    pos = plot[1][][idx]
    # project to screen space and shift it to be correct on the root scene
    proj_pos = shift_project(scene, to_ndim(Point3f, pos, 0))
    # anchor the tooltip at the projected position
    update_tooltip_alignment!(inspector, proj_pos)

    # Update the final text of the tooltip.
    if haskey(plot, :inspector_label)
        tt.text[] = plot[:inspector_label][](plot, idx, pos)
    else
        tt.text[] = position2string(pos)
    end
    # Show the tooltip
    tt.visible[] = true

    # return true to indicate that we have updated the tooltip
    return true
end

Next we want to mark the rectangle we are hovering. In this case we can use the rectangles which BarPlot passes to Poly, i.e. plot.plots[1][1][][idx]. The DataInspector contains some functionality for keeping track of temporary plots, so we can plot the indicator to the same scene that BarPlot uses. Doing so results in

using Makie:
    parent_scene, shift_project, update_tooltip_alignment!, position2string,
    clear_temporary_plots!

function show_data(inspector::DataInspector, plot::BarPlot, idx)
    # inspector.attributes holds some attributes relevant to indicators and is
    # used as a cache for indicator observables
    a = inspector.attributes
    tt = inspector.plot
    scene = parent_scene(plot)

    pos = plot[1][][idx]
    proj_pos = shift_project(scene, plot, to_ndim(Point3f, pos, 0))
    update_tooltip_alignment!(inspector, proj_pos)

    # We only want to mark the rectangle if that setting is enabled
    if a.enable_indicators[]
        # Get the relevant rectangle
        bbox = plot.plots[1][1][][idx]

        # If we haven't yet created an indicator create it
        if inspector.selection != plot
            # clear old indicators
            clear_temporary_plots!(inspector, plot)

            # Create the new indicator using some settings from `DataInspector`.
            p = wireframe!(
                scene, bbox, model = plot.model[], color = a.indicator_color,
                strokewidth = a.indicator_linewidth, linestyle = a.indicator_linestyle,
                visible = a.indicator_visible, inspectable = false
            )

            # tooltips are pushed forward a certain amount to make sure they're
            # drown on top of other things. This indicator should also be pushed
            # forward that much
            translate!(p, Vec3f(0, 0, a.depth[]))

            # Keep track of the indicator plot
            push!(inspector.temp_plots, p)

        # If we have already created an indicator plot we just need to update 
        # it. In this case we only need to update the rectangle.
        elseif !isempty(inspector.temp_plots)
            p = inspector.temp_plots[1]
            p[1][] = bbox
        end

        # Moving away from a plot will automatically set this to false, so we 
        # always need to set it to true.
        a.indicator_visible[] = true
    end

    if haskey(plot, :inspector_label)
        tt.text[] = plot[:inspector_label][](plot, idx, pos)
    else
        tt.text[] = position2string(pos)
    end
    tt.visible[] = true

    return true
end

which finishes the implementation of a custom tooltip for BarPlot.

Per-plot show_data

It is also possible to replace a call to show_data on a per-plot basis via the inspector_hover attribute. DataInspector assumes this to be a function (inspector, this_plot, index, hovered_child) -> Bool. You can also set up custom clean up with plot.inspector_clear = (inspector, plot) -> ... which is called whenever the plot is deselected.