cos-comparison 0.1.4

An AGI-spurting work (experimental stage now) based on local similarity comparison for feature extraction – biologically inspired, zero‑training edge / pattern detection.

Cos Comparison

Local similarity comparison for feature extraction – biologically inspired, zero‑training edge / pattern detection.

Core Idea

Information is produced by local comparison in raw data.
This module implements the centre‑surround antagonism mechanism from neuroscience, extracting edges, textures, and keypoints using only sliding window similarity.

The main formula (cosine‑modulated similarity):

$$ \text{cosmod} = \frac{2,(A\cdot B)}{|A|^2 + |B|^2} $$

• Spurt for real AGI
• No training, no labels, no backpropagation
• Works on 1D, 2D, 3D, 4D data (audio, images, video, volumes)
• Supports passive (reflex) and active (template matching) modes
• Pure Python core + optional NumPy / C acceleration

Core Principles

1. Information is generated by local comparison – edges, textures, patterns arise from comparing neighbouring regions.
2. Centre‑surround antagonism – two sliding windows compared with a fixed displacement vector d, mimicking retinal ganglion cells.
3. Three complementary similarity measures – cos (angular), mod (magnitude), cosmod (recommended combination).
4. Dual operational modes – passive (detects boundaries without templates) and active (template matching with user‑supplied kernel).
5. Multi‑scale and multi‑directionality – vary window size for scale, change d for orientation selectivity (vertical, horizontal, diagonal).
6. Dimension‑agnostic – same algorithm works on 1D, 2D, 3D, 4D and beyond.
7. Zero training, zero labels – deterministic computation, ready to use.
8. Full determinism and interpretability – every output has a clear geometric meaning.
9. Modular, pluggable architecture – pure Python reference, NumPy vectorised, C high‑performance backends with automatic fallback.

See core_base.txt for the complete design philosophy.

⚠️ Warning

This is an experimental version!

• Do not use it in a productive environment.
• All interfaces may change during the experimental stage, although I will try to keep it as stable as possible.
• Do not share this software with others without including this warning.

Roadmap & How to Contribute

This is still an early‑stage project. You can:

• Fix bugs and report issues.
• Compare with other algorithms (deep learning, classical filters) on standard benchmarks.
• Test on diverse datasets – time series, medical images, audio, video.
• Develop the core ideas – propose new similarity metrics, add GPU backends, or integrate into larger pipelines.

If you encounter import errors after installation, try modifying the package's __init__.py:

from cos_comparison import *

Installation

pip install cos-comparison

Quick Start

import cos_comparison as cc

# 1D passive: find similar segments
data = [1.0, 2.0, 3.0, 4.0, 5.0]
result = cc.cos_comparison_passive_1d(data, window_size=(2,), step=(1,), d=(1,))
print(result)

# 2D passive: edge detection
image = [[1,1,0,0], [1,1,0,0], [0,0,1,1], [0,0,1,1]]
edges = cc.cos_comparison_passive_2d(image, window_size=(2,2), step=(1,1), d=(1,0))
print(edges)

# Active mode – template matching
kernel = [1.0, 0.0]
result_active = cc.cos_comparison_active_1d(data, kernel=kernel, step=(1,))
print(result_active)

# Whole‑tensor cosine
a, b = [1,2,3], [2,3,4]
sim = cc.cos_1d(a, b)
print(sim)

Optional Backends

The package automatically selects the fastest available backend:

Priority	Backend	Requirement
1	C	Compile cos_comparison_c/include/core.c
2	NumPy	pip install numpy
3	Pure Python	Works out of the box

Force a specific backend via environment variable:

export COS_BACKEND=cos_comparison,cos_comparison_numpy

API Overview

Passive mode (self‑similarity)

cos_comparison_passive_1d / _2d / _3d / _4d + generic N‑dim

Active mode (template matching)

cos_comparison_active_1d / _2d / _3d / _4d + generic N‑dim

Whole‑tensor cosine

cos_1d / _2d / _3d / _4d

Statistics

mean_local_* / local_variance_* – sliding window mean and variance

Command line

python -m cos_comparison passive --data input.json --window 3 3 --output out.json

License

MIT © 2025 Li Jinxin. See LICENSE for details.