Basic Dash Callbacks

This is the 3rd chapter of the Dash Tutorial.
The previous chapter covered the Dash app layout
and the next chapter covers interactive graphing.
Just getting started? Make sure to install the necessary dependencies.

In the previous chapter we learned that app.layout describes what the app looks like and is a hierarchical tree of components.
The DashHtmlComponents library provides classes for all of the HTML tags, and the keyword arguments describe the HTML attributes like style, className, and id.
The DashCoreComponents library generates higher-level components like controls and graphs.

This chapter describes how to make your Dash apps using callback functions: functions that are automatically called by Dash whenever an input component’s property changes, in order to update some property in another component (the output).

For optimum user-interaction and chart loading performance, production
Dash apps should consider the Job Queue,
HPC, Datashader,
and horizontal scaling capabilities of Dash Enterprise.

Let’s get started with a simple example of an interactive Dash app.

Simple Interactive Dash App

If you’re using Dash Enterprise’s Data Science Workspaces,
copy & paste the below code into your Workspace (see video).

Find out if your company is using Dash Enterprise

using Dash

app = dash()

app.layout = html_div() do
    html_h6("Change the value in the text box to see callbacks in action!"),
    html_div(
        children = [
            "Input: ",
            dcc_input(id = "my-input", value = "initial value", type = "text")
        ],
    ),
    html_br(),
    html_div(id = "my-output")
end

callback!(app, Output("my-output", "children"), Input("my-input", "value")) do input_value
    "Output: $(input_value)"
end

run_server(app, "0.0.0.0", debug=true)
Change the value in the text box to see callbacks in action!
Input:

Let’s break down this example:

  1. The “inputs” and “outputs” of our application are described
    as the arguments of the callback! function definition.

  2. In Dash, the inputs and outputs of our application are simply the
    properties of a particular component. In this example,
    our input is the “value” property of the component that has the ID
    my-input”. Our output is the “children” property of the
    component with the ID “my-output”.

  3. Whenever an input property changes, the function that the
    callback decorator wraps will get called automatically.
    Dash provides this callback function with the new value of the input property as
    its argument, and Dash updates the property of the output component
    with whatever was returned by the function.
  4. The component_id and component_property keywords are optional
    (there are only two arguments for each of those objects).
    They are included in this example for clarity but will be omitted in the rest of the documentation for the sake of brevity and readability.
  5. Don’t confuse the Input object and the dcc_input object. The former is just used in these callback definitions and the latter is an actual component.
  6. Notice how we don’t set a value for the children property of the
    my-output component in the layout. When the Dash app starts, it
    automatically calls all of the callbacks with the initial values of the
    input components in order to populate the initial state of the output
    components. In this example, if you specified the div component as
    html_div(id='my-output', children='Hello world'),
    it would get overwritten when the app starts.

It’s sort of like programming with Microsoft Excel:
whenever a cell changes (the input), all the cells that depend on that cell (the outputs)
will get updated automatically. This is called “Reactive Programming” because the outputs react to changes in the inputs automatically.

Remember how every component is described entirely through its
set of keyword arguments? Those arguments that we set in
Julia become properties of the component,
and these properties are important now.
With Dash’s interactivity, we can dynamically update any of those properties
using callbacks. Often we’ll update the children property of HTML
components to display new text (remember that children is responsible for the contents of a component) or the figure property of a dcc_graph
component to display new data. We could also update the style of a
component or even the available options of a dcc_dropdown component!


Let’s take a look at another example where a dcc_slider updates
a dcc_graph.

Dash App Layout With Figure and Slider

using Dash
using DataFrames, PlotlyJS, UrlDownload

df = DataFrame(urldownload("https://raw.githubusercontent.com/plotly/datasets/master/gapminderDataFiveYear.csv"))

years = unique(df[!, :year])

app = dash()

app.layout = html_div() do
    dcc_graph(id = "graph"),
    dcc_slider(
        id = "year-slider-1",
        min = minimum(years),
        max = maximum(years),
        marks = Dict([Symbol(v) => Symbol(v) for v in years]),
        value = minimum(years),
        step = nothing,
    )
end

callback!(
    app,
    Output("graph", "figure"),
    Input("year-slider-1", "value"),
) do selected_year
    return Plot(
        df[df.year .== selected_year, :],
        Layout(
            xaxis_type = "log",
            xaxis_title = "GDP Per Capita",
            yaxis_title = "Life Expectancy",
            legend_x = 0,
            legend_y = 1,
            hovermode = "closest",
            transition_duration = 500
        ),
        x = :gdpPercap,
        y = :lifeExp,
        text = :country,
        group = :continent,
        mode = "markers",
        marker_size = 15,
        marker_line_color = "white",
    )
end

run_server(app, "0.0.0.0", debug = true)

Theming with Dash Enterprise Design Kit

Default Theme
Default Theme

Mars Theme
Mars Theme

Neptune Theme
Neptune Theme

Miller Theme
Miller Theme

Extrasolar Theme
Extrasolar Theme

Preset Themes
Preset Themes

In this example, the "value" property of the dcc_slider is the
input of the app, and the output of the app is the "figure" property of the
dcc_graph.
Whenever the value of the dcc_slider changes, Dash calls the
callback function update_figure with the new value. The function filters the
dataframe with this new value, constructs a figure object,
and returns it to the Dash application.

There are a few nice patterns in this example:

  1. We
    load our dataframe at the start of the app:
    df = DataFrame(...).
    This dataframe df is in the global state of the app and can be
    read inside the callback functions.
  2. Loading data into memory can be expensive. By loading querying data at
    the start of the app instead of inside the callback functions, we ensure
    that this operation is only done once – when the app server starts. When a user
    visits the app or interacts with the app, that data (df) is already in memory.
    If possible, expensive initialization (like downloading or querying
    data) should be done in the global scope of the app instead of within
    the callback functions.
  3. The callback does not modify the original data, it only creates copies
    of the dataframe by filtering .
    This is important: your callbacks should never modify variables
    outside of their scope
    . If your callbacks modify global state, then one
    user’s session might affect the next user’s session and when the app is
    deployed on multiple processes or threads, those modifications will not
    be shared across sessions.
  4. We are turning on transitions with layout.transition to give an idea
    of how the dataset evolves with time: transitions allow the chart to
    update from one state to the next smoothly, as if it were animated.

Dash App With Multiple Inputs

In Dash, any “output” can have multiple “input” components.
Here’s a simple example that binds five inputs
(the value property of two dcc_dropdown components,
two dcc_radioitems components, and one dcc_slider component)
to one output component (the figure property of the dcc_graph component).
Notice how callback! lists all five Input items after the Output.

using Dash
using DataFrames, PlotlyJS, UrlDownload

df2 = DataFrame(urldownload("https://raw.githubusercontent.com/plotly/datasets/master/country_indicators.csv"))

dropmissing!(df2)

rename!(df2, Dict(:"Year" => "year"))

available_indicators = unique(df2[!, "Indicator Name"])
years = unique(df2[!, "year"])

app = dash()

app.layout = html_div() do
    html_div(
        children = [
            dcc_dropdown(
                id = "xaxis-column",
                options = [
                    (label = i, value = i) for i in available_indicators
                ],
                value = "Fertility rate, total (births per woman)",
            ),
            dcc_radioitems(
                id = "xaxis-type",
                options = [(label = i, value = i) for i in ["linear", "log"]],
                value = "linear",
            ),
        ],
        style = (width = "48%", display = "inline-block"),
    ),
    html_div(
        children = [
            dcc_dropdown(
                id = "yaxis-column",
                options = [
                    (label = i, value = i) for i in available_indicators
                ],
                value = "Life expectancy at birth, total (years)",
            ),
            dcc_radioitems(
                id = "yaxis-type",
                options = [(label = i, value = i) for i in ["linear", "log"]],
                value = "linear",
            ),
        ],
        style = (width = "48%", display = "inline-block", float = "right"),
    ),
    dcc_graph(id = "indicator-graphic"),
    dcc_slider(
        id = "year-slider-2",
        min = minimum(years),
        max = maximum(years),
        marks = Dict([Symbol(v) => Symbol(v) for v in years]),
        value = minimum(years),
        step = nothing,
    )
end

callback!(
    app,
    Output("indicator-graphic", "figure"),
    Input("xaxis-column", "value"),
    Input("yaxis-column", "value"),
    Input("xaxis-type", "value"),
    Input("yaxis-type", "value"),
    Input("year-slider-2", "value"),
) do xaxis_column_name, yaxis_column_name, xaxis_type, yaxis_type, year_value
    df2f = df2[df2.year .== year_value, :]
    return Plot(
        df2f[df2f[!, Symbol("Indicator Name")] .== xaxis_column_name, :Value],
        df2f[df2f[!, Symbol("Indicator Name")] .== yaxis_column_name, :Value],
        Layout(
            xaxis_type = xaxis_type == "Linear" ? "linear" : "log",
            xaxis_title = xaxis_column_name,
            yaxis_title = yaxis_column_name,
            yaxis_type = yaxis_type == "Linear" ? "linear" : "log",
            hovermode = "closest",
        ),
        kind = "scatter",
        text = df2f[
            df2f[!, Symbol("Indicator Name")] .== yaxis_column_name,
            Symbol("Country Name"),
        ],
        mode = "markers",
        marker_size = 15,
        marker_opacity = 0.5,
        marker_line_width = 0.5,
        marker_line_color = "white"
    )
end

run_server(app, "0.0.0.0", debug = true)

Theming with Dash Enterprise Design Kit

Default Theme
Default Theme

Mars Theme
Mars Theme

Neptune Theme
Neptune Theme

Miller Theme
Miller Theme

Extrasolar Theme
Extrasolar Theme

Design Kit Theme Editor
Design Kit Theme Editor

In this example, the callback executes whenever the value property of any of the
dcc_dropdown, dcc_slider,
or dcc_radioitems components change.

The input arguments of the callback are the current
value of each of the “input” properties, in the order that they were
specified.

Even though only a single Input changes at a time (i.e. a user can only change
the value of a single Dropdown in a given moment), Dash collects the
current state of all the specified Input properties and passes them
into the callback function. These callback functions are always guaranteed
to receive the updated state of the app.

Let’s extend our example to include multiple outputs.

Dash App With Multiple Outputs

So far all the callbacks we’ve written only update a single Output property.
We can also update several outputs at once: list all the properties you want to update
in callback!,
and return that many items from the callback. This is particularly useful if
two outputs depend on the same computationally intensive intermediate result,
such as a slow database query.

using Dash

app = dash()

app.layout = html_div() do
    dcc_input(id = "input-4", value = "1", type = "text"),
    html_tr((html_td("x^2 ="), html_td(id = "square"))),
    html_tr((html_td("x^3 ="), html_td(id = "cube"))),
    html_tr((html_td("2^x ="), html_td(id = "twos"))),
    html_tr((html_td("3^x ="), html_td(id = "threes"))),
    html_tr((html_td("x^x ="), html_td(id = "xx")))
end

callback!(
    app,
    Output("square", "children"),
    Output("cube", "children"),
    Output("twos", "children"),
    Output("threes", "children"),
    Output("xx", "children"),
    Input("input-4", "value"),
) do x
    if x == "" || x == nothing
        return ("", "", "", "", "")
    end

    x = parse(Int64, x)
    return (x^2, x^3, 2^x, 3^x, x^x)
end

run_server(app, "0.0.0.0", debug=true)
x 2
x 3
2 x
3 x
x x

A word of caution: it’s not always a good idea to combine outputs, even if
you can:

Dash App With Chained Callbacks

You can also chain outputs and inputs together: the output of one callback
function could be the input of another callback function.

This pattern can be used to create dynamic UIs where, for example, one input component
updates the available options of another input component.
Here’s a simple example.

using Dash
using CSV, DataFrames

app = dash()

all_options = Dict(
    "America" => ["New York City", "San Francisco", "Cincinnati"],
    "Canada" => ["Montreal", "Toronto", "Ottawa"],
)

app.layout = html_div() do
    html_div(
        children = [
            dcc_radioitems(
                id = "countries-radio",
                options = [(label = i, value = i) for i in keys(all_options)],
                value = "America",
            ),
            html_hr(),
            dcc_radioitems(id = "cities-radio"),
            html_hr(),
            html_div(id = "display-selected-values"),
        ],
    )
end

callback!(
    app,
    Output("cities-radio", "options"),
    Input("countries-radio", "value"),
) do selected_country
    return [(label = i, value = i) for i in all_options[selected_country]]
end

callback!(
    app,
    Output("cities-radio", "value"),
    Input("cities-radio", "options"),
) do available_options
    return available_options[1][:value]
end

callback!(
    app,
    Output("display-selected-values", "children"),
    Input("countries-radio", "value"),
    Input("cities-radio", "value"),
) do selected_country, selected_city
    return "$(selected_city) is a city in $(selected_country) "
end

run_server(app, "0.0.0.0", debug=true)


The first callback updates the available options in the second
dcc_radioitems component based off of the selected value in the
first dcc_radioitems component.

The second callback sets an initial value when the options property
changes: it sets it to the first value in that options array.

The final callback displays the selected value of each component.
If you change the value of the countries dcc_radioitems
component, Dash will wait until the value of the cities component is updated
before calling the final callback. This prevents your callbacks from being
called with inconsistent state like with "America" and "Montréal".

Dash App With State

In some cases, you might have a “form”-like pattern in your
application. In such a situation, you may want to read the value
of an input component, but only when the user is finished
entering all of their information in the form rather than immediately after
it changes.

Attaching a callback to the input values directly can look like this:

using Dash

app = dash()

app.layout = html_div() do
    dcc_input(id = "input-1", type = "text", value = "Montreal"),
    dcc_input(id = "input-2", type = "text", value = "Canada"),
    html_div(id = "output-keywords")
end

callback!(
    app,
    Output("output-keywords", "children"),
    Input("input-1", "value"),
    Input("input-2", "value"),
) do input_1, input_2
    return "Input 1 is \"$input_1\" and Input 2 is \"$input_2\""
end

run_server(app, "0.0.0.0", debug=true)

In this example, the callback function is fired whenever any of the
attributes described by the Input change.
Try it for yourself by entering data in the inputs above.

State allows you to pass along extra values without
firing the callbacks. Here’s the same example as above but with the two
dcc_input components as State
and a new button component as an Input.

using Dash

app = dash()

app.layout = html_div() do
    dcc_input(id = "input-1-state", type = "text", value = "Montreal"),
    dcc_input(id = "input-2-state", type = "text", value = "Canada"),
    html_button(id = "submit-button-state", children = "submit", n_clicks = 0),
    html_div(id = "output-state")
end

callback!(
    app,
    Output("output-state", "children"),
    Input("submit-button-state", "n_clicks"),
    State("input-1-state", "value"),
    State("input-2-state", "value"),
) do clicks, input_1, input_2
    return "The Button has been pressed \"$clicks\" times, Input 1 is \"$input_1\" and Input 2 is \"$input_2\""
end

run_server(app, "0.0.0.0", debug=true)

In this example, changing text in the dcc_input boxes won’t fire
the callback, but clicking on the button will. The current values of the
dcc_input values are still passed into the callback even though
they don’t trigger the callback function itself.

Note that we’re triggering the callback by listening to the n_clicks property
of the html_button component. n_clicks is a property that gets
incremented every time the component has been clicked on.
It’s available in every component in
the DashHtmlComponents library, but most useful with buttons.

Summary

We’ve covered the fundamentals of callbacks in Dash.
Dash apps are built off of a set
of simple but powerful principles: UIs that are customizable
through reactive callbacks.
Every attribute/property of a component can be modified
as the output of a callback, while a subset of the attributes (such as the value
property of dcc_dropdown component)
are editable by the user through interacting with the page.


The next part of the Dash tutorial covers interactive graphing. Dash Tutorial Part 4: Interactive Graphing