Interactive Charts#

One of the unique features of Altair, inherited from Vega-Lite, is a declarative grammar of not just visualization, but also interaction. This is both convenient and powerful, as we will see in this section. There are three core concepts of this grammar:

  • Parameters are the basic building blocks in the grammar of interaction. They can either be simple variables or more complex selections that map user input (e.g., mouse clicks and drags) to data queries.

  • Conditions and filters can respond to changes in parameter values and update chart elements based on that input.

  • Widgets and other chart input elements can bind to parameters so that charts can be manipulated via drop-down menus, radio buttons, sliders, legends, etc.

Parameters#

Parameters are the building blocks of interaction in Altair. There are two types of parameters: variables and selections. We introduce these concepts through a series of examples.

Note

This material was changed considerably with the release of Altair 5.

Variables: Reusing Values#

Variable parameters allow for a value to be defined once and then reused throughout the rest of the chart. Here is a simple scatter-plot created from the cars dataset:

import altair as alt
from vega_datasets import data

cars = data.cars.url

alt.Chart(cars).mark_circle().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N'
)

Variable parameters are created using the param() function. Here, we create a parameter with a default value of 0.1 using the value property:

op_var = alt.param(value=0.1)

In order to use this variable in the chart specification, we explicitly add it to the chart using the add_params() method, and we can then reference the variable within the chart specification. Here we set the opacity using our op_var parameter.

op_var = alt.param(value=0.1)

alt.Chart(cars).mark_circle(opacity=op_var).encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N'
).add_params(
    op_var
)

It’s reasonable to ask whether all this effort is necessary. Here is a more natural way to accomplish the same thing that avoids the use of both param() and add_params.

op_var2 = 0.1

alt.Chart(cars).mark_circle(opacity=op_var2).encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N'
)

The benefit of using param() doesn’t become apparent until we incorporate an additional component. In the following example we use the bind property of the parameter, so that the parameter becomes bound to an input element. In this example, that input element is a slider widget.

slider = alt.binding_range(min=0, max=1, step=0.05, name='opacity:')
op_var = alt.param(value=0.1, bind=slider)

alt.Chart(cars).mark_circle(opacity=op_var).encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N'
).add_params(
    op_var
)

Now we can dynamically change the opacity of the points in our chart using the slider. You will learn much more about binding parameters to input elements such as widgets in the section Bindings & Widgets.

Note

A noteworthy aspect of Altair’s interactivity is that these effects are controlled entirely within the web browser. This means that you can save charts as HTML files and share them with your colleagues who can access the interactivity via their browser without the need to install Python.

Selections: Capturing Chart Interactions#

Selection parameters define data queries that are driven by interactive manipulation of the chart by the user (e.g., via mouse clicks or drags). There are two types of selections: selection_interval() and selection_point().

Here we will create a simple chart and then add an selection interval to it. We could create a selection interval via param(select="interval"), but it is more convenient to use the shorter selection_interval.

Here is a simple scatter-plot created from the cars dataset:

import altair as alt
from vega_datasets import data

cars = data.cars.url

alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N'
)

First we’ll create an interval selection using the selection_interval() function (an interval selection is also referred to as a “brush”):

brush = alt.selection_interval()

We can now add this selection interval to our chart via add_params:

alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N'
).add_params(
    brush
)

The result above is a chart that allows you to click and drag to create a selection region, and to move this region once the region is created.

So far this example is very similar to what we did in the variable example: we created a selection parameter using brush = alt.selection_interval(), and we attached that parameter to the chart using add_params. One difference is that here we have not defined how the chart should respond to the selection; you will learn this in the next section.

Conditions & Filters#

Conditional Encodings#

The example above is neat, but the selection interval doesn’t actually do anything yet. To make the chart respond to this selection, we need to reference the selection in within the chart specification. Here, we will use the condition() function to create a conditional color encoding: we’ll tie the color to the "Origin" column for points in the selection, and set the color to "lightgray" for points outside the selection:

alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color=alt.condition(brush, 'Origin:N', alt.value('lightgray'))
).add_params(
    brush
)

As you can see, the color of the points now changes depending on whether they are inside or outside the selection. Above we are using the selection parameter brush as a predicate (something that evaluates as True or False). This is controlled by the line color=alt.condition(brush, 'Origin:N', alt.value('lightgray')). Data points which fall within the selection evaluate as True, and data points which fall outside the selection evaluate to False. The 'Origin:N' specifies how to color the points which fall within the selection, and the alt.value('lightgray') specifies that the outside points should be given a constant color value; you can remember this as alt.condition(<condition>, <if_true>, <if_false>).

This approach becomes even more powerful when the selection behavior is tied across multiple views of the data within a compound chart. For example, here we create a chart object using the same code as above, and horizontally concatenate two versions of this chart: one with the x-encoding tied to "Horsepower", and one with the x-encoding tied to "Acceleration"

chart = alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color=alt.condition(brush, 'Origin:N', alt.value('lightgray'))
).properties(
    width=250,
    height=250
).add_params(
    brush
)

chart | chart.encode(x='Acceleration:Q')

Because both copies of the chart reference the same selection object, the renderer ties the selections together across panels, leading to a dynamic display that helps you gain insight into the relationships within the dataset.

Each selection type has attributes through which its behavior can be customized; for example we might wish for our brush to be tied only to the "x" encoding to emphasize that feature in the data. We can modify the brush definition, and leave the rest of the code unchanged:

brush = alt.selection_interval(encodings=['x'])

chart = alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color=alt.condition(brush, 'Origin:N', alt.value('lightgray'))
).properties(
    width=250,
    height=250
).add_params(
    brush
)

chart | chart.encode(x='Acceleration:Q')

Filtering Data#

Using a selection parameter to filter data works in much the same way as using it within condition, For example, in transform_filter(brush), we are again using the selection parameter brush as a predicate. Data points which evaluate to True (i.e., data points which lie within the selection) are kept, and data points which evaluate to False are filtered out.

It is not possible to both select and filter in the same chart, so typically this functionality will be used when at least two sub-charts are present. In the following example, we attach the selection parameter to the upper chart, and then filter data in the lower chart based on the selection in the upper chart. You can explore how the counts change in the bar chart depending on the size and position of the selection in the scatter plot.

brush = alt.selection_interval()

points = alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N'
).add_params(
    brush
)

bars = alt.Chart(cars).mark_bar().encode(
    x='count()',
    y='Origin:N',
    color='Origin:N'
).transform_filter(
    brush
)

points & bars

Selection Types#

Now that we have seen the basics of how we can use a selection to interact with a chart, let’s take a more systematic look at some of the types of selection parameters available in Altair. For simplicity, we’ll use a common chart in all the following examples; a simple heat-map based on the cars dataset. For convenience, let’s write a quick Python function that will take a selection object and create a chart with the color of the chart elements linked to this selection:

def make_example(selector):
    cars = data.cars.url

    return alt.Chart(cars).mark_rect().encode(
        x="Cylinders:O",
        y="Origin:N",
        color=alt.condition(selector, 'count()', alt.value('lightgray'))
    ).properties(
        width=300,
        height=180
    ).add_params(
        selector
    )

Next we’ll use this function to demonstrate the properties of various selections.

Interval Selections#

An interval selection allows you to select chart elements by clicking and dragging. You can create such a selection using the selection_interval() function:

interval = alt.selection_interval()
make_example(interval)

As you click and drag on the plot, you’ll find that your mouse creates a box that can be subsequently moved to change the selection.

The selection_interval() function takes a few additional arguments; for example we can bind the interval to only the x-axis, and set it such that the empty selection contains none of the points:

interval_x = alt.selection_interval(encodings=['x'], empty=False)
make_example(interval_x)

Point Selections#

A point selection allows you to select chart elements one at a time via mouse actions. By default, points are selected on click:

point = alt.selection_point()
make_example(point)

By changing some arguments, we can select points on mouseover rather than on click. We can also set the nearest flag to True so that the nearest point is highlighted:

point_nearest = alt.selection_point(on='mouseover', nearest=True)
make_example(point_nearest)

Point selections also allow for multiple chart objects to be selected. By default, chart elements can be added to and removed from the selection by clicking on them while holding the shift key, you can try in the two charts above.

Selection Targets#

For any but the simplest selections, the user needs to think about exactly what is targeted by the selection, and this can be controlled with either the fields or encodings arguments. These control what data properties are used to determine which points are part of the selection.

For example, here we create a small chart that acts as an interactive legend, by targeting the Origin field using fields=['Origin']. Clicking on points in the upper-right plot (the legend) will propagate a selection for all points with a matching Origin.

selection = alt.selection_point(fields=['Origin'])
color = alt.condition(
    selection,
    alt.Color('Origin:N').legend(None),
    alt.value('lightgray')
)

scatter = alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color=color,
    tooltip='Name:N'
)

legend = alt.Chart(cars).mark_point().encode(
    alt.Y('Origin:N').axis(orient='right'),
    color=color
).add_params(
    selection
)

scatter | legend

The above could be equivalently replace fields=['Origin'] with encodings=['color'], because in this case the chart maps color to 'Origin'. Also note that there is a shortcut to create interactive legends in Altair described in the section Legend Binding.

Similarly, we can specify multiple fields and/or encodings that must be matched in order for a datum to be included in a selection. For example, we could modify the above chart to create a two-dimensional clickable legend that will select points by both Origin and number of cylinders:

selection = alt.selection_point(fields=['Origin', 'Cylinders'])
color = alt.condition(
    selection,
    alt.Color('Origin:N').legend(None),
    alt.value('lightgray')
)

scatter = alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color=color,
    tooltip='Name:N'
)

legend = alt.Chart(cars).mark_rect().encode(
    alt.Y('Origin:N').axis(orient='right'),
    x='Cylinders:O',
    color=color
).add_params(
    selection
)

scatter | legend

By fine-tuning the behavior of selections in this way, they can be used to create a wide variety of linked interactive chart types.

Parameter Composition#

Altair also supports combining multiple parameters using the &, | and ~ for respectively AND, OR and NOT logical composition operands.

Returning to our heatmap examples, we can construct a scenario where there are two people who can make an interval selection in the same chart. The person Alex makes a selection box when the alt-key (macOS: option-key) is selected and Morgan can make a selection box when the shift-key is selected. We use the Brushconfig to give the selection box of Morgan a different style. Now, we color the rectangles when they fall within Alex’s or Morgan’s selection (note that you need to create both selections before seeing the effect).

alex = alt.selection_interval(
    on="[mousedown[event.altKey], mouseup] > mousemove",
    name='alex'
)
morgan = alt.selection_interval(
    on="[mousedown[event.shiftKey], mouseup] > mousemove",
    mark=alt.BrushConfig(fill="#fdbb84", fillOpacity=0.5, stroke="#e34a33"),
    name='morgan'
)

alt.Chart(cars).mark_rect().encode(
    x='Cylinders:O',
    y='Origin:O',
    color=alt.condition(alex | morgan, 'count()', alt.ColorValue("grey"))
).add_params(
    alex, morgan
).properties(
    width=300,
    height=180
)

With these operators, selections can be combined in arbitrary ways:

  • ~(alex & morgan): to select the rectangles that fall outside Alex’s and Morgan’s selections.

  • alex | ~morgan: to select the rectangles that fall within Alex’s selection or outside the selection of Morgan

Bindings & Widgets#

With an understanding of the parameter types and conditions, you can now bind parameters to chart elements (e.g. legends) and widgets (e.g. drop-downs and sliders). This is done using the bind option inside param and selection. As specified by the Vega-lite binding docs, there are three types of bindings available:

  1. Point and interval selections can be used for data-driven interactive elements, such as highlighting and filtering based on values in the data.

  2. Sliders and checkboxes can be used for logic-driven interactive elements, such as highlighting and filtering based on the absolute values in these widgets.

  3. Interval selections can be bound to a scale, such as zooming in on a map.

The following table summarizes the input elements that are supported in Vega-Lite:

Input Element

Description

Example

binding_checkbox

Renders as checkboxes allowing for multiple selections of items.

Multiple Interactions

binding_radio

Radio buttons that force only a single selection

Multiple Interactions

binding_select

Drop down box for selecting a single item from a list

Multiple Interactions

binding_range

Shown as a slider to allow for selection along a scale.

US Population by Age and Sex

binding

General method that supports many HTML input elements

Widget Binding#

Widgets are HTML input elements, such as drop-downs, sliders, radio buttons, and search boxes. There are a three strategies for how variable and selection parameters can be used together with widgets: data-driven lookups, data-driven comparisons, and logic-driven comparisons.

Data-Driven Lookups#

Data-driven lookups use the active value(s) of the widget together with a selection parameter to look up points with matching values in the chart’s dataset. For example, we can establish a binding between an input widget and a point selection to filter the data as in the example below where a drop-down is used to highlight cars of a specific Origin:

input_dropdown = alt.binding_select(options=['Europe','Japan','USA'], name='Region ')
selection = alt.selection_point(fields=['Origin'], bind=input_dropdown)
color = alt.condition(
    selection,
    alt.Color('Origin:N').legend(None),
    alt.value('lightgray')
)

alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color=color,
).add_params(
    selection
)

Note that although it looks like a value is selected in the dropdown from the start, we need to set value= to actually start out with an initial selection in the chart. We did this previously with variable parameters and selection parameters follow the same pattern as you will see further down in the Encoding Channel Binding section.

As you can see above, we are still using conditions to make the chart respond to the selection, just as we did without widgets. Bindings and input elements can also be used to filter data allowing the user to see just the selected points as in the example below. In this example, we also add an empty selection to illustrate how to revert to showing all points after a selection has been made in a radio button or drop-down (which cannot be deselected).

# Make radio button less cramped by adding a space after each label
# The spacing will only show up in your IDE, not on this doc page
options = ['Europe', 'Japan', 'USA']
labels = [option + ' ' for option in options]

input_dropdown = alt.binding_radio(
    # Add the empty selection which shows all when clicked
    options=options + [None],
    labels=labels + ['All'],
    name='Region: '
)
selection = alt.selection_point(
    fields=['Origin'],
    bind=input_dropdown,
)

alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    # We need to set a constant domain to preserve the colors
    # when only one region is shown at a time
    color=alt.Color('Origin:N').scale(domain=options),
).add_params(
    selection
).transform_filter(
    selection
)

In addition to the widgets listed in the table above, Altair has access to any html widget via the more general binding function. In the example below, we use a search input to filter points that match the search string exactly. You can hover over the points to see the car names and try typing one into the search box, e.g. vw pickup to see the point highlighted (you need to type out the full name).

search_input = alt.selection_point(
    fields=['Name'],
    empty=False,  # Start with no points selected
    bind=alt.binding(
        input='search',
        placeholder="Car model",
        name='Search ',
    )
)
alt.Chart(data.cars.url).mark_point(size=60).encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    tooltip='Name:N',
    opacity=alt.condition(
        search_input,
        alt.value(1),
        alt.value(0.05)
    )
).add_params(
    search_input
)

It is not always useful to require an exact match to the search syntax, and when we will be learning about Expressions, we will see how we can match partial strings via a regex instead.

Data-Driven Comparisons#

So far we have seen the use of selections to lookup points with precisely matching values in our data. This is often useful, but sometimes we might want to make a more complex comparison than an exact match. For example, we might want to create a condition we select the points in the data that are above or below a threshold value, which is specified via a slider. For this workflow it is recommended to use variable parameters via param and as you can see below, we use the special syntax datum.xval to reference the column to compare against. Prefixing the column name with datum tells Altair that we want to compare to a column in the dataframe, rather than to a Python variable called xval, which would have been the case if we just wrote xval < selector.

import numpy as np
import pandas as pd


rand = np.random.RandomState(42)
df = pd.DataFrame({
    'xval': range(100),
    'yval': rand.randn(100).cumsum()
})

slider = alt.binding_range(min=0, max=100, step=1, name='Cutoff ')
selector = alt.param(name='SelectorName', value=50, bind=slider)

alt.Chart(df).mark_point().encode(
   x='xval',
   y='yval',
   color=alt.condition(
       alt.datum.xval < selector,
       # 'datum.xval < SelectorName',  # An equivalent alternative
       alt.value('red'),
       alt.value('blue')
   )
).add_params(
   selector
)

In this particular case we could actually have used a selection parameter since selection values can be accessed directly and used in expressions that affect the chart. For example, here we create a slider to choose a cutoff value, and color points based on whether they are smaller or larger than the value:

slider = alt.binding_range(min=0, max=100, step=1, name='Cutoff ')
selector = alt.selection_point(
    name="SelectorName",
    fields=['cutoff'],
    bind=slider,
    value=[{'cutoff': 50}]
)

alt.Chart(df).mark_point().encode(
    x='xval',
    y='yval',
    color=alt.condition(
        alt.datum.xval < selector.cutoff,
        # 'datum.xval < SelectorName.cutoff',  # An equivalent alternative
        alt.value('red'),
        alt.value('blue')
    )
).add_params(
    selector
)

While it can be useful to know how to access selection values in expression strings, using the parameters syntax introduced in Altair 5 often provides a more convenient syntax for simple interactions like this one since they can also be accessed in expression strings as we saw above. Similarly, it is often possible to use equality statements such as alt.datum.xval == selector to lookup exact values but it is often more convenient to switch to a selection parameter and specify a field/encoding.

Logic-Driven Comparisons#

A logic comparison is a type of comparison that is based on logical rules and conditions, rather than on the actual data values themselves. For example, for a checkbox widget we want to check if the state of the checkbox is True or False and execute some action depending on whether it is checked or not. When we are using a checkbox as a toggle like this, we need to use param instead of selection_point, since we don’t want to check if there are True/False values in our data, just if the value of the check box is True (checked) or False (unchecked):

bind_checkbox = alt.binding_checkbox(name='Scale point size by "Acceleration": ')
param_checkbox = alt.param(bind=bind_checkbox)

alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    size=alt.condition(
        param_checkbox,
        'Acceleration:Q',
        alt.value(25)
    )
).add_params(
    param_checkbox
)

Another example of creating a widget binding that is independent of the data, involves an interesting use case for the more general binding function. In the next example, this function introduces a color picker where the user can choose the colors of the chart interactively:

color_usa = alt.param(value="#317bb4", bind=alt.binding(input='color', name='USA '))
color_europe = alt.param(value="#ffb54d", bind=alt.binding(input='color', name='Europe '))
color_japan = alt.param(value="#adadad", bind=alt.binding(input='color', name='Japan '))

alt.Chart(data.cars.url).mark_circle().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color=alt.Color(
        'Origin:N',
        scale=alt.Scale(
            domain=['USA', 'Europe', 'Japan'],
            range=[color_usa, color_europe, color_japan]
        )
    )
).add_params(
    color_usa, color_europe, color_japan
)

Encoding Channel Binding#

There is no direct way to map an encoding channel to a widget in order to dynamically display different charts based on different column choices, such as y=column_param. The underlying reason this is not possible is that in Vega-Lite, the field property does not accept a parameter as value; see the field Vega-Lite documentation. You can follow the discussion in this issue vega/vega-lite#7365, and in the meantime, you can use parameters for a convenient workaround which let’s you achieve the same functionality and change the plotted columns based on a widget selection (the x-axis title cannot be changed dynamically, but a text mark could be used instead if desired):

dropdown = alt.binding_select(
    options=['Horsepower', 'Displacement', 'Weight_in_lbs', 'Acceleration'],
    name='X-axis column '
)
xcol_param = alt.param(
    value='Horsepower',
    bind=dropdown
)

alt.Chart(data.cars.url).mark_circle().encode(
    x=alt.X('x:Q').title(''),
    y='Miles_per_Gallon:Q',
    color='Origin:N'
).transform_calculate(
    x=f'datum[{xcol_param.name}]'
).add_params(
    xcol_param
)

It was possible to achieve something similar before the introduction of parameters in Altair 5 by using transform_fold and transform_filter, but the spec for this is more complex (as can be seen in this SO answer) so the solution above is to prefer.

Legend Binding#

An interactive legend can often be helpful to assist in focusing in on groups of data. Instead of manually having to build a separate chart to use as a legend, Altair provides the bind='legend' option to facilitate the creation of clickable legends:

selection = alt.selection_point(fields=['Origin'], bind='legend')

alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N',
    opacity=alt.condition(selection, alt.value(0.8), alt.value(0.2))
).add_params(
    selection
)

Scale Binding#

With interval selections, the bind property can be set to the value of "scales". In these cases, the binding will automatically respond to the panning and zooming along the chart:

selection = alt.selection_interval(bind='scales')

alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N',
).add_params(
    selection
)

Because this is such a common pattern, Altair provides the interactive() method which creates a scale-bound selection more concisely:

alt.Chart(cars).mark_point().encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    color='Origin:N',
).interactive()

Expressions#

Altair allows custom interactions by utilizing the expression language of Vega for writing basic formulas. A Vega expression string is a well-defined set of JavaScript-style operations. To simplify building these expressions in Python, Altair provides the expr module, which offers constants and functions to construct expressions using Python syntax. Both JavaScript-syntax and Python-syntax are supported within Altair to define an expression and an introductory example of each is available in the Calculate transform documentation so we recommend checking out that page before continuing.

In the following example, we define a range connected to a parameter named param_width. We then assign two expressions via param using both JavaScript and Python-syntax. Using these two expressions defined as a parameter, we can connect them to an encoding channel option, such as the title color of the axis. If the width is below 200, then the color is red; otherwise, the color is blue.

bind_range = alt.binding_range(min=100, max=300, name='Slider value:  ')
param_width = alt.param(bind=bind_range)

# Examples of how to write both js and python expressions
param_color_js_expr = alt.param(expr=f"{param_width.name} < 200 ? 'red' : 'black'")
param_color_py_expr = alt.param(expr=alt.expr.if_(param_width < 200, 'red', 'black'))

chart = alt.Chart(df).mark_point().encode(
    alt.X('xval').axis(titleColor=param_color_js_expr),
    alt.Y('yval').axis(titleColor=param_color_py_expr)
).add_params(
    param_width,
    param_color_js_expr,
    param_color_py_expr
)
chart

In the example above, we used a JavaScript-style ternary operator f"{param_width.name} < 200 ? 'red' : 'blue'" which is equivalent to the Python function expr.if_(param_width < 200, 'red', 'blue'). The expressions defined as parameters also needed to be added to the chart within .add_params().

In addition to assigning an expression within a parameter definition as shown above, the expr() utility function allows us to define expressions inline, add_params. In the next example, we modify the chart above to change the size of the points based on an inline expression. Instead of creating a conditional statement, we use the value of the expression as the size directly and therefore only need to specify the name of the parameter.

chart.mark_point(size=alt.expr(param_width.name))

Inline expressions defined by expr(...) are not parameters so they can be added directly in the chart spec instead of via add_params.

Another option to include an expression within a chart specification is as a value definition to an encoding channel. Here, we make the exact same modification to the chart as in the previous example via this alternate approach:

chart.encode(size=alt.value(alt.expr(param_width.name)))

Some parameter names have special meaning in Vega-Lite, for example, naming a parameter width will automatically link it to the width of the chart. In the example below, we also modify the chart title to show the value of the parameter:

bind_range = alt.binding_range(min=100, max=300, name='Chart width: ')
param_width = alt.param('width', bind=bind_range)

# In Javascript, a number is converted to a string when added to an existing string,
# which is why we use this nested quotation.
title=alt.Title(alt.expr(f'"This chart is " + {param_width.name} + " px wide"'))
alt.Chart(df, title=title).mark_point().encode(
    alt.X('xval'),
    alt.Y('yval')
).add_params(
    param_width,
)

Now that we know the basics of expressions, let’s see how we can improve on our search input example and make the search string match via a regex pattern. To do this we need to use expr.regex to define the regex string, and expr.test to test it against another string (in this case the string in the Name column). The i option makes the regex case insensitive, and you can see that we have switched to using param instead of selection_point since we are doing something more complex than looking up values with an exact match in the data. To try this out, you can type mazda|ford in the search input box below.

search_input = alt.param(
    value='',
    bind=alt.binding(
        input='search',
        placeholder="Car model",
        name='Search ',
    )
)
alt.Chart(data.cars.url).mark_point(size=60).encode(
    x='Horsepower:Q',
    y='Miles_per_Gallon:Q',
    tooltip='Name:N',
    opacity=alt.condition(
        alt.expr.test(alt.expr.regexp(search_input, 'i'), alt.datum.Name),
        # f"test(regexp({search_input.name}, 'i'), datum.Name)",  # Equivalent js alternative
        alt.value(1),
        alt.value(0.05)
    )
).add_params(
    search_input
)

And remember, all this interactivity is client side. You can save this chart as an HTML file or put it on a static site generator such as GitHub/GitLab pages and anyone can interact with it without having to install Python. Quite powerful!

To summarize expressions:

  • Altair can utilize the expression language of Vega for writing basic formulas to enable custom interactions.

  • Both JavaScript-style syntax and Python-style syntax are supported in Altair to define expressions.

  • Altair provides the expr module which allows expressions to be constructed with Python syntax.

  • Expressions can be included within a chart specification using two approaches: through a param(expr=...) parameter definition or inline using the expr(...) utility function.

  • Expressions can be used anywhere the documentation mentions that an ExprRef is an accepted value. This is mainly in three locations within a chart specification: mark properties, encoding channel options, and within a value definition for an encoding channel. They are also supported in the chart title, but not yet for subtitles or guide titles (i.e. axis and legends, see vega/vega-lite#7408 for details).

Further Examples#

Now that you understand the basics of Altair selections and bindings, you might wish to look through the Interactive Charts section of the example gallery for ideas about how they can be applied to more interesting charts.

For more information on how to fine-tune selections, including specifying other mouse and keystroke options, see the Vega-Lite Selection documentation.

Limitations#

Some possible use cases for the above interactivity are not currently supported by Vega-Lite, and hence are not currently supported by Altair. Here are some examples.

  1. If we are using a selection_point, it would be natural to want to return information about the chosen data point, and then process that information using Python. This is not currently possible, so any data processing will have to be handled using tools such as transform_calculate, etc. You can follow the progress on this in the following issue: altair-viz/altair#1153.

    • The dashboarding package Panel has added support for processing Altair selections with custom callbacks in their 0.13 release. This is currently the only Python dashboarding package that supports custom callbacks for Altair selections and you can read more about how to use this functionality in the Panel documentation.