Introduction
Bokeh is a Python library for interactive browser-based visualizations — hover, zoom, pan, linked brushing, and real-time streaming. Streamlit renders Bokeh charts with st.bokeh_chart. This post covers the patterns that matter.
Setup
pip install streamlit bokeh pandas numpy
Step 1: Basic line chart
import streamlit as st
from bokeh.plotting import figure
import numpy as np
x = np.linspace(0, 10, 100)
y = np.sin(x)
p = figure(title="Sine Wave", x_axis_label="x", y_axis_label="y", width=700, height=400)
p.line(x, y, legend_label="sin(x)", line_width=2, color="#3b82f6")
st.bokeh_chart(p, use_container_width=True)
Step 2: Scatter with hover
from bokeh.models import HoverTool
p = figure(title="Scatter", width=700, height=400)
p.scatter("x", "y", size=8, color="category", source=data, legend_field="category", alpha=0.7)
hover = HoverTool(tooltips=[("X", "@x"), ("Y", "@y"), ("Category", "@category")])
p.add_tools(hover)
st.bokeh_chart(p, use_container_width=True)
Step 3: Bar chart
from bokeh.transform import factor_cmap
from bokeh.palettes import Spectral6
fruits = ["Apples", "Pears", "Nectarines", "Plums", "Grapes", "Strawberries"]
counts = [5, 3, 4, 2, 4, 6]
p = figure(x_range=fruits, height=400, title="Fruit Counts")
p.vbar(x=fruits, top=counts, width=0.9, color=factor_cmap("x", palette=Spectral6, factors=fruits))
p.xgrid.grid_line_color = None
p.y_range.start = 0
st.bokeh_chart(p, use_container_width=True)
Step 4: Linked brushing
from bokeh.layouts import gridplot
from bokeh.models import ColumnDataSource
source = ColumnDataSource(data)
p1 = figure(width=400, height=300, title="Plot 1")
p1.circle("x", "y", source=source, size=8, color="navy", alpha=0.5)
p2 = figure(width=400, height=300, title="Plot 2")
p2.circle("x", "z", source=source, size=8, color="firebrick", alpha=0.5)
p1.x_range = p2.x_range
st.bokeh_chart(gridplot([[p1, p2]]), use_container_width=True)
Selecting points in one plot highlights them in the other.
Step 5: Real-time streaming
from bokeh.models import ColumnDataSource
import time
source = ColumnDataSource(data=dict(x=[], y=[]))
p = figure(width=700, height=400, x_range=(0, 100))
p.line("x", "y", source=source)
placeholder = st.empty()
for i in range(100):
source.stream({"x": [i], "y": [np.sin(i * 0.1)]}, rollover=50)
placeholder.bokeh_chart(p, use_container_width=True)
time.sleep(0.1)
source.stream() appends data with rollover limiting buffer size.
Step 6: JS callbacks (no server round-trip)
from bokeh.models import Slider, CustomJS
slider = Slider(start=0, end=10, value=1, step=0.1, title="Frequency")
p = figure(width=700, height=400)
r = p.line(x, np.sin(x), line_width=2)
callback = CustomJS(args=dict(line=r, slider=slider), code="""
const freq = slider.value;
const data = line.data_source.data;
for (let i = 0; i < data.x.length; i++) data.y[i] = Math.sin(data.x[i] * freq);
line.data_source.change.emit();
""")
slider.js_on_change("value", callback)
st.bokeh_chart(slider, use_container_width=True)
st.bokeh_chart(p, use_container_width=True)
JS callbacks run in the browser — instant updates without server round-trips.
Bokeh vs Plotly vs Altair vs Matplotlib
| Bokeh | Plotly | Altair | Matplotlib | |
|---|---|---|---|---|
| Interactivity | High (JS) | High | Medium | None |
| Linked brushing | Yes | No | Yes | No |
| Streaming | Yes | Limited | No | No |
| Learning curve | Medium | Medium | Medium | Low |
| Best for | Advanced interactivity | Dashboards | Statistical viz | Publications |
Common mistakes
- Not using
use_container_width=True— charts won’t resize. - Too many glyphs — 10k+ points lag. Use
ColumnDataSourcewith sampling. - Forgetting
p.xgrid.grid_line_color = None— default grid is noisy. - Mixing Bokeh versions — 3.x has breaking changes from 2.x. Pin your version.
- Not closing figures — Bokeh figures are lightweight; no cleanup needed, but avoid creating them in loops without reusing.
Key Concepts
- Bokeh = browser-based interactive visualization (Leaflet for charts)
figure()= create a plotColumnDataSource= shared data for linked plotsHoverTool= tooltips on hoverstream()= append data for real-timeCustomJS= client-side callbacks (no server round-trip)
What to Try Next
- Build a real-time dashboard with Bokeh streaming.
- Use linked brushing for data exploration.
- Combine with caching for large datasets.
- Export charts with
bokeh.io.export_png.
FAQ
Bokeh vs Plotly — which is better?
Bokeh = more control over interactivity (linked brushing, streaming, JS callbacks). Plotly = easier to use, more chart types, better for dashboards. Use Bokeh for advanced interactivity; Plotly for quick dashboards.
Can Bokeh handle 100k+ points?
Not smoothly in the browser. For large datasets, use Datashader (renders to image) or switch to server-side rendering. Bokeh is best for <10k points.
Is Bokeh free?
Yes — open source (BSD license). Rendering happens client-side (BokehJS), no server cost. Works offline.