Bokeh Tutorial

03. Adding Annotations

In [1]:
from import output_notebook, show
from bokeh.plotting import figure
In [2]:
Loading BokehJS ...


Sometimes we want to add visual cues (boundary lines, shaded regions, labels and arrows, etc.) to our plots to call out some feature or other. Bokeh has several annotation types available for uses like this. Typically to add annotations we create the "low level" annotation object directly, and add it to our Plot, Figure or Chart using add_layout. Let's take a look at some specific examples.


Spans are "infinite" vertical or horizonal lines. When creating them, you specify the dimension that should be spanned (i.e., width or height), any visual line properties for the appearance, and the location along the dimension where the line should be drawn. Let's look at an example that adds two horizontal spans to a simple plot:

In [3]:
import numpy as np
from bokeh.models.annotations import Span

x = np.linspace(0, 20, 200)
y = np.sin(x)

p = figure(y_range=(-2, 2))
p.line(x, y)

upper = Span(location=1, dimension='width', line_color='olive', line_width=4)

lower = Span(location=-1, dimension='width', line_color='firebrick', line_width=4)

In [4]:
# Exercise:

Box Annotations

Sometimes you might want to call out some region of the plot by drawing a shaded box. This can be done with the BoxAnnotation, which is configured with the coordinate properties:

  • top
  • left
  • bottom
  • right

as well as any visual line or fill properties to control the appearance.

"Infinite" boxes can be made by leaving any of the coordinates unspecified. E.g., if top is not given, the box will always extend to the top of the plot area, regardless of any panning or zooming that happens.

Let's take a look at an example that adds a few shaded boxes to a plot:

In [5]:
import numpy as np
from bokeh.models.annotations import BoxAnnotation

x = np.linspace(0, 20, 200)
y = np.sin(x)

p = figure(y_range=(-2, 2))
p.line(x, y)

# region that always fills the top of the plot
upper = BoxAnnotation(bottom=1, fill_alpha=0.1, fill_color='olive')

# region that always fills the bottom of the plot
lower = BoxAnnotation(top=-1, fill_alpha=0.1, fill_color='firebrick')

# a finite region
center = BoxAnnotation(top=0.6, bottom=-0.3, left=7, right=12, fill_alpha=0.1, fill_color='navy')

In [6]:
# Exercise:


The Label annotation allows you to easily attach single text labels to plots. The position and text to display are configured as x, y, and text:

Label(x=10, y=5, text="Some Label")

By default the units are in "data space" but x_units and y_units maybe set to "screen" to position the label relative to the canvas. Labels can also accept x_offset and y_offset to offset the final position from x and y by a given screen space distance.

Label objects also have standard text, line (border_line) and fill (background_fill) properties. The line and fill properties apply to a bounding box around the text:

Label(x=10, y=5, text="Some Label", text_font_size="12pt", 
      border_line_color="red", background_fill_color="blue")
In [7]:
from bokeh.models.annotations import Label
from bokeh.plotting import figure

p = figure(x_range=(0,10), y_range=(0,10))[2, 5, 8], [4, 7, 6], color="olive", size=10)

label = Label(x=5, y=7, x_offset=12, text="Second Point", text_baseline="middle")

In [8]:
# EXERCISE: experiment with Label


The LabelSet annotation allows you to create many labels at once, for instance if you want to label an entire set of scatter markers. They are similar to Label, but they can also accept a ColumnDataSource as the source property, and then x and y may refer to columns in the data source, e.g. x="col2" (but may also still be fixed values, e.g. x=10).

In [9]:
from bokeh.plotting import figure
from bokeh.models import ColumnDataSource, LabelSet

source = ColumnDataSource(data=dict(
    temp=[166, 171, 172, 168, 174, 162],
    pressure=[165, 189, 220, 141, 260, 174],
    names=['A', 'B', 'C', 'D', 'E', 'F']))

p = figure(x_range=(160, 175))
p.scatter(x='temp', y='pressure', size=8, source=source)
p.xaxis.axis_label = 'Temperature (C)'
p.yaxis.axis_label = 'Pressure (lbs)'

labels = LabelSet(x='temp', y='pressure', text='names', level='glyph',
                  x_offset=5, y_offset=5, source=source, render_mode='canvas')


In [10]:
# EXERCISE: experiment with LabelSet


The Arrow annotation allows you to "point" at different things on your plot, and can be especially useful in conjuction with labels.

For example, to create an arrow that points from (0,0) to (1,1):

p.add_layout(Arrow(x_start=0, y_start=0, x_end=1, y_end=0))

This arrow will have the default OpenHead arrow head at the end of the arrow. Other kinds of arrow heads include NormalHead and VeeHead. The arrow head type can be controlled by setting the start and end properties of Arrow objects:

p.add_layout(Arrow(start=OpenHead(), end=VeeHead(), 
             x_start=0, y_start=0, x_end=1, y_end=0))

This will create a double-ended arrow with an "open" head at the start, and a "vee" head at the end. Arrowheads have the standard set of line and fill properties to control their appearance. As an example

OpenHead(line_color="firebrick", line_width=4)

The code and plot below shows several of these configurations together.

In [11]:
from bokeh.models.annotations import Arrow
from bokeh.models.arrow_heads import OpenHead, NormalHead, VeeHead

p = figure(plot_width=600, plot_height=600)[0, 1, 0.5], y=[0, 0, 0.7], radius=0.1,
         color=["navy", "yellow", "red"], fill_alpha=0.1)

p.add_layout(Arrow(end=OpenHead(line_color="firebrick", line_width=4),
                   x_start=0, y_start=0, x_end=1, y_end=0))

                   x_start=1, y_start=0, x_end=0.5, y_end=0.7))

p.add_layout(Arrow(end=VeeHead(size=35), line_color="red",
                   x_start=0.5, y_start=0.7, x_end=0, y_end=0))

In [12]:
# EXERCISE: experiment with arrows and arrow heads


Simple Legends

In [13]:
import numpy as np

x = np.linspace(0, 4*np.pi, 100)
y = np.sin(x)

p = figure(height=400), y, legend="sin(x)")
p.line(x, 2*y, legend="2*sin(x)", line_dash=[4, 4], line_color="orange", line_width=2)


Compound legends

Sometimes, two (or more) different glyphs are used with a single data source. In this case, you can make compound legends by specifying the same legend argument to multiple glyph methods when creating a plot, as in:, y, legend="sin(x)")
p.line(x, y, legend="sin(x)", line_dash=[4, 4], line_color="orange", line_width=2)
In [14]:
# (1) Try making a compound legend 
# (2) Try moving the legend using p.legend.location.  Possible values are listed in bokeh.core.enum

Legend Orientation and Positioning

In [15]:
# Exercise

Color bars

In [16]:
from bokeh.sampledata.autompg import autompg
from bokeh.models import LinearColorMapper, ColorBar
from bokeh.palettes import Viridis256

source = ColumnDataSource(autompg)
color_mapper = LinearColorMapper(palette=Viridis256, low=autompg.weight.min(), high=autompg.weight.max())

p = figure(x_axis_label='Year', y_axis_label='MPG', tools='', toolbar_location='above')'yr', y='mpg', color={'field': 'weight', 'transform': color_mapper}, size=20, alpha=0.6, source=source)

color_bar = ColorBar(color_mapper=color_mapper, label_standoff=12, location=(0,0), title='Weight')
p.add_layout(color_bar, 'right')