fastquadtree 0.4.0

Rust-accelerated quadtree for Python with fast inserts, range queries, and k-NN search.

fastquadtree

Rust-optimized quadtree with a simple Python API.

• Python package: fastquadtree
• Python ≥ 3.8
• Import path: from fastquadtree import QuadTree

Install

pip install fastquadtree

If you are developing locally:

# optimized dev install
maturin develop --release

Quickstart

from fastquadtree import QuadTree

# Bounds are (min_x, min_y, max_x, max_y)
qt = QuadTree(bounds=(0, 0, 1000, 1000), capacity=20)  # max_depth is optional

# Insert points with auto ids
id1 = qt.insert((10, 10))
id2 = qt.insert((200, 300))
id3 = qt.insert((999, 500), id=42)  # you can supply your own id

# Axis-aligned rectangle query
hits = qt.query((0, 0, 250, 350))  # returns [(id, x, y), ...] by default
print(hits)  # e.g. [(1, 10.0, 10.0), (2, 200.0, 300.0)]

# Nearest neighbor
best = qt.nearest_neighbor((210, 310))  # -> (id, x, y) or None
print(best)

# k-nearest neighbors
top3 = qt.nearest_neighbors((210, 310), 3)
print(top3)  # list of up to 3 (id, x, y) tuples

# Delete items by ID and location
deleted = qt.delete(id2, (200, 300))  # True if found and deleted
print(f"Deleted: {deleted}")
print(f"Remaining items: {qt.count_items()}")

# For object tracking with track_objects=True
qt_tracked = QuadTree((0, 0, 1000, 1000), capacity=4, track_objects=True)
player1 = {"name": "Alice", "score": 100}
player2 = {"name": "Bob", "score": 200}

id1 = qt_tracked.insert((50, 50), obj=player1)
id2 = qt_tracked.insert((150, 150), obj=player2)

# Delete by object reference (O(1) lookup!)
deleted = qt_tracked.delete_by_object(player1, (50, 50))
print(f"Deleted player: {deleted}")  # True

Working with Python objects

You can keep the tree pure and manage your own id → object map, or let the wrapper manage it.

Option A: Manage your own map

from fastquadtree import QuadTree

qt = QuadTree((0, 0, 1000, 1000), capacity=16)
objects: dict[int, object] = {}

def add(obj) -> int:
    obj_id = qt.insert(obj.position)  # auto id
    objects[obj_id] = obj
    return obj_id

# Later, resolve ids back to objects
ids = [obj_id for (obj_id, x, y) in qt.query((100, 100, 300, 300))]
selected = [objects[i] for i in ids]

Option B: Ask the wrapper to track objects

from fastquadtree import QuadTree

qt = QuadTree((0, 0, 1000, 1000), capacity=16, track_objects=True)

# Store the object alongside the point
qt.insert((25, 40), obj={"name": "apple"})

# Ask for Item objects so you can access .obj lazily
items = qt.query((0, 0, 100, 100), as_items=True)
for it in items:
    print(it.id, it.x, it.y, it.obj)

You can also attach or replace an object later:

qt.attach(123, my_object)  # binds object to id 123

API

QuadTree(bounds, capacity, *, max_depth=None, track_objects=False, start_id=1)

• bounds — tuple (min_x, min_y, max_x, max_y) defines the 2D area covered by the quadtree
• capacity — max number of points kept in a leaf before splitting
• max_depth — optional depth cap. If omitted, the tree can keep splitting as needed
• track_objects — if True, the wrapper maintains an id → object map for convenience.
• start_id — starting value for auto-assigned ids

Core Methods

Full docs are in the docstrings of the Python Shim

• insert(xy, *, id=None, obj=None) -> int

• insert_many_points(points) -> int

• query(rect, *, as_items=False) -> list

• nearest_neighbor(xy, *, as_item=False) -> (id, x, y) | Item | None

• nearest_neighbors(xy, k, *, as_items=False) -> list

• delete(id, xy) -> bool

• delete_by_object(obj) -> bool (requires track_objects=True)

• attach(id, obj) -> None (requires track_objects=True)

• count_items() -> int

• get(id) -> object | None

• get_all_rectangles() -> list[tuple] (for visualization)

Item (returned when as_items=True)

• Attributes: id, x, y, and a lazy obj property
• Accessing obj performs a dictionary lookup only if tracking is enabled

Geometric conventions

• Rectangles are (min_x, min_y, max_x, max_y).
• Containment rule is closed on the min edge and open on the max edge (x >= min_x and x < max_x and y >= min_y and y < max_y). This only matters for points exactly on edges.

Performance tips

• Choose capacity so that leaves keep a small batch of points. Typical values are 8 to 64.
• If your data is very skewed, set a max_depth to prevent long chains.
• For fastest local runs, use maturin develop --release.
• The wrapper keeps Python overhead low: raw tuple results by default, Item wrappers only when requested.

Benchmarks

fastquadtree outperforms all other quadtree python packages (at least all the ones I could find and install via pip.)

Library comparison

Summary (largest dataset, PyQtree baseline)

• Points: 500,000, Queries: 500
• Fastest total: fastquadtree at 2.207 s

Library	Build (s)	Query (s)	Total (s)	Speed vs PyQtree
fastquadtree	0.321	1.885	2.207	4.27×
Rtree	1.718	4.376	6.095	1.55×
nontree-QuadTree	1.617	7.643	9.260	1.02×
PyQtree	4.349	5.082	9.431	1.00×
quads	3.874	9.058	12.932	0.73×
e-pyquadtree	2.732	10.598	13.330	0.71×
Brute force	0.019	19.986	20.005	0.47×

Native vs Shim

Setup

• Points: 500,000
• Queries: 500
• Repeats: 5

Timing (seconds)

Variant	Build	Query	Total
Native	0.483	4.380	4.863
Shim (no map)	0.668	4.167	4.835
Shim (track+objs)	1.153	4.458	5.610

Overhead vs Native

• No map: build 1.38x, query 0.95x, total 0.99x
• Track + objs: build 2.39x, query 1.02x, total 1.15x

Run benchmarks

To run the benchmarks yourself, first install the dependencies:

pip install -r benchmarks/requirements.txt

Then run:

python benchmarks/cross_library_bench.py
python benchmarks/benchmark_native_vs_shim.py 

Run Visualizer

A visualizer is included to help you understand how the quadtree subdivides space.

pip install -r interactive/requirements.txt
python interactive/interactive_v2.py

Check the CLI arguments for the cross-library benchmark in benchmarks/quadtree_bench/main.py.

FAQ

What happens if I insert the same id more than once? Allowed. For k-nearest, duplicates are de-duplicated by id. For range queries you will see every inserted point.

Can I delete items from the quadtree? Yes! Use delete(id, xy) to remove specific items. You must provide both the ID and exact location for precise deletion. This handles cases where multiple items exist at the same location. If you're using track_objects=True, you can also use delete_by_object(obj, xy) for convenient object-based deletion with O(1) lookup. The tree automatically merges nodes when item counts drop below capacity.

Can I store rectangles or circles? The core stores points. To index objects with extent, insert whatever representative point you choose. For rectangles you can insert centers or build an AABB tree separately.

Threading Use one tree per thread if you need heavy parallel inserts from Python.

License

MIT. See LICENSE.

Acknowledgments

• Python libraries compared: PyQtree, e-pyquadtree, Rtree, nontree, quads
• Built with PyO3 and maturin