mannr

mannr is a lightweight, single-file Python package that implements a randomized approximate nearest neighbor (ANN) tree for tabular data using Pandas and NumPy. It focuses on simplicity and offline experimentation rather than production-grade performance.

The package recursively partitions data using randomly sampled hyperplanes, producing a binary tree whose leaves contain small subsets of the original dataset.

Features

• Single-file Python module (no package directory)
• Randomized ANN tree construction Works directly with Pandas DataFrames
• Supports arbitrary numeric feature subsets
• Simple tree and node abstraction

Installation

Clone the repository and install locally:

pip install .

Or install in editable mode for development:

pip install -e .

Since this is a single-file module, installation uses py_modules instead of packages.

Dependencies

The core functionality requires:

• Python 3.8+
• pandas
• numpy

Project Structure

.
├── mannr.py
├── setup.py
└── README.md

Core Concepts

ANN Tree Construction

The tree is built recursively using a random hyperplane split:

1. Two rows are randomly sampled from the dataset.

2. Their midpoint defines a center.

3. The vector between them defines a hyperplane.

4. All points are split based on which side of the hyperplane they lie on.

5. Recursion continues until the subset size is ≤ k.

This approach is inspired by randomized ANN methods such as RP-trees.

Notes

A node is considered a leaf if at least one child is missing.

Traversal uses a queue (collections.deque).

Intended for debugging or exploration, not analysis

Example Usage

import pandas as pd
from mannr import gen_ann_tree, find_leaves

df = pd.read_csv("data.csv")
features = ["x", "y"]

root = gen_ann_tree(df, features, k=20)
leaves = find_leaves(root)