A pure-Python layout algorithm, not a charting library. Give it values sorted largest-first and a rectangle to fill; it returns tiles whose area is proportional to value, shaped as close to squares as it can. squarify.plot() draws them in Matplotlib — or take the raw {x, y, dx, dy} dicts and render them anywhere.
Distilled & cross-checked against the squarify source of truth: github.com/laserson/squarify (README + __init__.py API docs) · PyPI · python-charts.com · python-graph-gallery.com · algorithm: Bruls, Huizing & van Wijk, "Squarified Treemaps" (2000)
import squarifyimport matplotlib.pyplot as pltvalues = [500, 433, 78, 25, 25, 7]values.sort(reverse=True) # REQUIRED: largest first# --- path A: one call draws it -------------------------------squarify.plot(sizes=values, label=["A","B","C","D","E","F"], pad=True, alpha=.8, ec="white") # ec → bar edgecolorplt.axis("off") # a treemap has no axes# --- path B: raw rectangles, render anywhere -----------------norm = squarify.normalize_sizes(values, 700, 433) # sum → 700*433rects = squarify.squarify(norm, 0, 0, 700, 433) # x, y, dx, dy# [{'x': 0.0, 'y': 0.0, 'dx': 327.7, 'dy': 433.0}, ...]$ pip install squarifyimport squarify★import matplotlib.pyplot as plt★squarify.plot().import pandas as pd# layout works without matplotlib installedsquarify.plot(sizes, ...)★squarify.normalize_sizes(sizes, dx, dy)★sum(sizes) == dx*dy.squarify.squarify(sizes, x, y, dx, dy)★squarify.padded_squarify(sizes, x, y, dx, dy)# that's it — no classes, no statesquarify.plot(sizes=values)★norm_x=100, norm_y=100pad=True★padded_squarify so tiles have gaps.ax=my_axax = squarify.plot(...)(fig, ax) pair.pad=0.25not a sizepad is a boolean switch; any truthy value gives the same fixed gap.values.sort(reverse=True)★sorted(values, reverse=True)# values must be positive★df = df.sort_values("v", ascending=False)# unsorted input still "works"silentsquarify.normalize_sizes(values, dx, dy)★norm = normalize_sizes(v, 700, 433)sum(norm) == 700 * 433.# proportions are preserved# plot() calls this internallysquarify(raw_values, ...)wrongrects = squarify.squarify(norm, 0, 0, 700, 433)★(x, y) plus width dx and height dy.padded_squarify(norm, 0, 0, 700, 433)# output order == input order★rects[i] pairs with values[i] and your labels.x=0, y=0# Bruls, Huizing & van Wijk (2000){"x": 0.0, "y": 0.0, "dx": 327.7, "dy": 433.0}★r["x"], r["y"]r["dx"] * r["dy"]json.dumps(rects)ax.add_patch(Rectangle((r["x"], r["y"]), r["dx"], r["dy"]))label=["A", "B", "C"]★sizes.value=valueslabel=[f"{n}\n{v}" for n, v in zip(names, values)]★\n gives a two-line name + value label.text_kwargs={"fontsize": 10, "color": "white"}★Axes.text.text_kwargs={"weight": "bold"}# labels overflow tiny tilesgotchacolor=["#91DCEA", "#64CDCC", "#5FBB68"]★color="steelblue"color=sns.color_palette("magma", len(values))★color=[cm.viridis(i/len(v)) for i in range(len(v))]alpha=0.8# default colours are RANDOMgotchacolor= for stable output.ec="white"★bar_kwargs={"edgecolor": "k", "linewidth": 2}★Axes.bar.linewidth=2bar_kwargs.# explicit kwargs win over bar_kwargspad=True, ec="white", linewidth=2plt.axis("off")★plt.figure(figsize=(12, 7))plt.title("Market share")fig, ax = plt.subplots() ; squarify.plot(..., ax=ax)★plt.savefig("tree.png", dpi=300, bbox_inches="tight")norm_x=200, norm_y=20g = df.groupby("cat")["sales"].sum()★g = g.sort_values(ascending=False)★squarify.plot(sizes=g.values, label=g.index)★df["pct"] = df.v / df.v.sum() * 100g = g.head(12)basic: plot(sizes=v) ; plt.axis("off")labelled: plot(sizes=v, label=names, value=v)crisp: plot(sizes=v, pad=True, ec="white", linewidth=2)palette: plot(sizes=v, color=sns.color_palette("rocket", len(v)), alpha=.9)to JSON: json.dumps(squarify(normalize_sizes(v,w,h),0,0,w,h))nested: squarify(sub, r["x"], r["y"], r["dx"], r["dy"])# flat only — no hierarchylimit# no interactivity / tooltipslimitplotly.express.treemap(...)plotly.express.sunburst(...)d3.treemap()# treemaps beat pies past ~6 slicesThe rule the whole package rests on, why "squarified" matters, the shape of what comes back, and the two ways to use it. The treemaps below are the algorithm's real output for the README's values [500, 433, 78, 25, 25, 7].
Every tile's area — not its width or height — encodes the value. This is the algorithm's actual output.
Naive slice-and-dice makes unreadable slivers. Squarify keeps every tile's aspect ratio as close to 1 as it can.
Each tile is a plain dict: an origin corner plus a width and a height. Nothing framework-specific.
One call draws a Matplotlib treemap; the layout functions hand back geometry you can render in anything.
normalize_sizes guarantees before layoutnormalize_sizespad=0.25 and pad=True are identicalcolor= or every run looks different(fig, ax) tuplerects[i] lines up with values[i] and your labelsbar_kwargs → Axes.barplotly.express.treemap