pyairtree

Python bindings for AirTree — a high-performance library for building, compressing, and querying multi-dimensional histograms over floating-point data.

Turn large numeric datasets into compact .airtree histograms and run fast statistical queries without keeping the full raw data in memory.

Required Notice: Copyright AirMettle, Inc. (https://airmettle.com)

---

Installation

pip install pyairtree

Supported platforms

Platform	Architecture	Python
Linux	x86_64, ARM	3.8 – 3.12
Windows	amd64	3.8 – 3.12

NumPy is installed automatically as a dependency.

Verify the install

import pyairtree
print(pyairtree.__version__)

---

Quick start

Build a 1D histogram from a NumPy array, save it, and query percentiles:

import numpy as np
import pyairtree

# Sample data
data = np.random.randn(100_000)

# Build a compressed histogram (1D, high precision)
buffer = pyairtree.generate(data)

# Save to disk
pyairtree.write(buffer, "histogram.airtree")

# Query percentiles
p = pyairtree.Percentile(buffer)
print(f"Median: {p.get_percentile(50):.4f}")
print(f"95th percentile: {p.get_percentile(95):.4f}")

Load data directly from .npy or single-array .npz files

buffer = pyairtree.generate("samples/data.npy")

---

Core workflow

A typical workflow has three steps:

1. Generate — compress numeric arrays into an in-memory histogram buffer
2. Store — write the buffer to a .airtree file
3. Query — run statistics or spatial queries on the buffer or file

import pyairtree

# 1. Generate
buffer = pyairtree.generate(x, y, options)

# 2. Store
pyairtree.write(buffer, "output.airtree")

# 3. Query
reader = pyairtree.read("output.airtree")
p = pyairtree.Percentile(buffer)

---

Generating histograms

Unified generate() API (recommended)

Pass one NumPy array per dimension. Use AirTreeOptions to control dimensionality and precision:

import numpy as np
import pyairtree

x = np.random.randn(50_000)
y = np.random.randn(50_000)

options = pyairtree.AirTreeOptions()
options.dimensions = 2
options.type = pyairtree.ConfigType.XP   # highest precision (default for most use cases)

buffer = pyairtree.generate(x, y, options)

Configuration types

Type	Precision	Best for
ConfigType.XT	13-bit	Large datasets, fastest builds
ConfigType.XF	16-bit	General-purpose balance
ConfigType.XP	20-bit (1D) / 10-bit (2D+)	Highest accuracy

Dimensionality

# 1D
buf_1d = pyairtree.generate(a, options)

# 2D
buf_2d = pyairtree.generate(a, b, options)

# 3D
buf_3d = pyairtree.generate(a, b, c, options)

# 4D
buf_4d = pyairtree.generate(a, b, c, d, options)

Set options.dimensions to match the number of arrays you pass.

Named generation functions

Convenience functions are also available for explicit configs:

arr = pyairtree.buildFPHArray(data)

# 1D
pyairtree.generate_1DxT(arr)   # 13-bit
pyairtree.generate_1DxF(arr)   # 16-bit
pyairtree.generate_1DxP(arr)   # 20-bit

# 2D / 3D / 4D
pyairtree.generate_2DxP(arr_x, arr_y)
pyairtree.generate_3DxP(arr_x, arr_y, arr_z)
pyairtree.generate_4DxP(arr_x, arr_y, arr_z, arr_w)

Supported array types

NumPy arrays with dtype float64, float32, int64, or int32.

---

File I/O

Write a histogram

pyairtree.write(buffer, "histogram.airtree")

Read a histogram

reader = pyairtree.read("histogram.airtree")

print(reader.dims)        # number of dimensions
print(reader.bit_length)  # encoding precision
print(reader.type)        # configuration type

header = reader.header
print(header.version)
print(header.nan_count)

---

Query operations

Percentile (1D)

p = pyairtree.Percentile(buffer)

median = p.get_percentile(50.0)
p95    = p.get_percentile(95.0)

Min / Max (1D)

mm = pyairtree.MinMax(buffer)
print(mm.get_min(), mm.get_max())

Top-K (1D)

Return the bins with the highest counts:

topk = pyairtree.TopK(buffer)
results = topk.top_k(10.0)   # top 10% of bins by count

for r in results:
    print(r.lower_bound, r.upper_bound, r.count)

CDF (1D)

cdf = pyairtree.CDF(buffer)
values = cdf.get_cdf()

Bounding box (2D)

Count values inside a rectangular region:

bbox_query = pyairtree.BoundingBox(buffer_2d)

region = pyairtree.BoundingBoxCoordinate2D(
    min_x=0.0, max_x=1.0,
    min_y=0.0, max_y=1.0,
)

safe, edge = bbox_query.get_counts(region)
(safe_coords, safe_count) = safe
(edge_coords, edge_count) = edge

Grid query (1D)

Evaluate the histogram over a regular grid:

gq = pyairtree.GridQuery(buffer)
result = gq.get_grid(spec)   # pass a GridSpec with min, max, steps
rows = result.materialize_rows()

---

Export

Export histogram data to common interchange formats:

pyairtree.export_tree(buffer, "output.arrow", pyairtree.ExportFormat.ARROW)
pyairtree.export_tree(buffer, "output.parquet", pyairtree.ExportFormat.PARQUET)
pyairtree.export_tree(buffer, "output.csv", pyairtree.ExportFormat.CSV)

---

Choosing a configuration

Use case	Suggested config
Exploratory analysis on large 1D datasets	ConfigType.XT or generate_1DxT
Production 1D analytics	ConfigType.XF or generate_1DxF
High-precision 1D requirements	ConfigType.XP or generate_1DxP
2D spatial / correlation data	generate_2DxP
3D / 4D multi-variate data	generate_3DxP / generate_4DxP

---

Error handling

import pyairtree
import numpy as np

try:
    pyairtree.generate(np.array(["not", "numeric"]))
except (TypeError, RuntimeError) as e:
    print(f"Generation failed: {e}")

try:
    pyairtree.Percentile(two_d_buffer)   # Percentile is 1D only
except RuntimeError as e:
    print(f"Query failed: {e}")

---

License

AirMettle Noncommercial License

This software is provided for noncommercial use only. Government use requires a separate commercial license.

• Personal, research, educational, and qualifying nonprofit use is permitted • Commercial or government use requires a license from AirMettle

For commercial licensing, support, or enterprise usage:

support@airmettle.com

Patent Pending — U.S. Patent Application Publication No. 2025/0217930 A1 (https://patents.google.com/patent/US20250217930A1)

See the bundled LICENSE file for full terms.

---

Links

• Homepage: https://airmettle.com
• Support: support@airmettle.com