Altair

Altair exposes a declarative grammar-of-graphics API that compiles to Vega-Lite - charts become JSON specs you can version, diff, and embed on the web.

Recipe

Quick-reference recipe card - copy-paste ready.

import altair as alt
import pandas as pd
 
df = pd.read_csv("sales.csv", parse_dates=["ordered_at"])
chart = (
    alt.Chart(df)
    .mark_line(point=True)
    .encode(x="ordered_at:T", y="sum(revenue):Q", color="region:N")
    .properties(width=600, height=300, title="Revenue by region")
)
chart.save("revenue.json")

When to reach for this:

• Specs you want in git as JSON, not imperative plotting code
• Layered charts (+, |) with readable composition
• Embedding in static sites via Vega-Embed
• Teaching visualization grammar without matplotlib boilerplate

Working Example

import altair as alt
import pandas as pd
import numpy as np
 
rng = np.random.default_rng(42)
df = pd.DataFrame(
    {
        "region": np.repeat(["East", "West"], 150),
        "revenue": rng.normal(150, 35, 300).clip(20),
        "units": rng.integers(1, 60, 300),
    }
)
 
base = alt.Chart(df).encode(
    x=alt.X("revenue:Q", bin=alt.Bin(maxbins=20), title="Revenue"),
    color="region:N",
)
 
hist = base.mark_bar(opacity=0.7).properties(width=400, height=250, title="Distribution")
 
scatter = (
    alt.Chart(df)
    .mark_circle(size=60, opacity=0.5)
    .encode(
        x="units:Q",
        y="revenue:Q",
        color="region:N",
        tooltip=["region", "units", "revenue"],
    )
    .properties(width=400, height=250, title="Units vs revenue")
)
 
dashboard = alt.hconcat(hist, scatter).resolve_scale(color="shared")
 
rule = (
    alt.Chart(pd.DataFrame({"y": [150]}))
    .mark_rule(color="firebrick", strokeDash=[4, 4])
    .encode(y="y:Q")
)
layered = scatter + rule
 
dashboard.save("dashboard.json")
layered.save("scatter_target.json")

What this demonstrates:

• Bin encoding in declarative form
• Horizontal concat with shared color scale
• Layered rule annotation with +
• JSON export for embedding

Deep Dive

How It Works

• Chart(data) + mark_* + encode channels (x, y, color, tooltip).
• Shorthand :Q quantitative, :N nominal, :T temporal, :O ordinal.
• Transforms (transform_filter, transform_aggregate) stay in spec.
• Compiler emits Vega-Lite JSON rendered by JS in browser or PNG via altair_saver.

Encoding Types

Python Notes

import altair as alt
 
alt.data_transformers.disable_max_rows()  # only when you accept large embeds
alt.renderers.enable("json")  # notebook default varies by version

Gotchas

• Default 5000 row limit - Altair samples large frames. Fix: aggregate in pandas/Polars first or disable max rows knowingly.
• Wide data - multiple y columns need transform_fold or melt. Fix: tidy to long format upstream.
• Sorting nominal axes - default alphabetical misorders months. Fix: sort=["Jan", "Feb", ...] or :O with order.
• Saving PNG without dependencies - needs vl-convert or altair_saver. Fix: ship JSON to front-end or install converter in CI.
• Interactive selections on huge data - slow client-side. Fix: filter data server-side before chart.

NzA2MjAyfDQ4Nm9pZ2N8b2kuc2VkaXVnZWRvYw==

Alternatives

FAQs

alt.Chart(df).mark_point().encode(x="x", y="y").facet("region:N")

alt.Chart(df).transform_filter(alt.datum.revenue > 100).mark_bar().encode(...)

rule = alt.Chart(df).mark_rule(strokeDash=[4,4]).encode(y="mean(revenue):Q")
chart + rule

.encode(color=alt.Color("region:N", scale=alt.Scale(scheme="set2")))

Related

• Plotly - interactive alternative
• Visualization Basics - chart choice
• Visualization Best Practices - accessible color
• Plotting Large Data - aggregate first

Stack versions: This page was written for Python 3.14.0 (stable 3.14, maintenance 3.13), FastAPI 0.115+, Django 5.2, Flask 3.1, Pydantic 2, PyTorch 2.6+, pandas 2.2+, Polars 1.x, ruff 0.9+, and uv 0.6+.