htm_py

A pure-Python, production-ready Hierarchical Temporal Memory (HTM) library.

Overview

htm_py implements key components of HTM theory:

• Encoders (RDSE and DateEncoder)
• Spatial Pooler
• Temporal Memory
• Anomaly Score
• Anomaly Likelihood
• Prediction Count

The library is engineered for:

• Faithfulness to Numenta’s NAB benchmark
• Python 3.x compatibility
• No C++ dependencies
• Lightweight, modular, and highly extensible

Installation

pip install -r requirements.txt

(Requirements are minimal — mainly numpy, scipy, and pytest.)

How to Run

Run NAB comparison:

python nab_tm_runner.py

Run full unit tests:

pytest tests

Repository Structure

htm_py/
    connections.py
    date_encoder.py
    rdse_encoder.py
    spatial_pooler.py
    temporal_memory.py
    htm_model.py
tests/
    test_connections.py
    test_temporal_memory.py
    test_encoders.py
    test_htm_model.py
data/
    art_daily_jumpsup.csv
nab_tm_runner.py
requirements.txt
pytest.ini
README.md

Notes

• htm_py matches Numenta NAB outputs to machine precision (anomaly scores & likelihoods).
• Designed to be easily extendable for production use cases.
• 100% Python, no need for htm.core, pycapnp, or old nupic bindings.

---

Built with ❤️ for robust, biologically inspired AI systems.