How wt-hu/pycausalfs works
Project type and purpose
• This is a pure‐Python open‐source library (not a web or client/server app) for causality‐based feature selection and Bayesian network learning.
• Its goal is to implement 30 representative causality‐based feature selection methods plus local/global BN structure learning, offering a common API to experiment, compare, extend, and integrate new algorithms.
File structure
Top‐level:
• README.md, manual‐pyCausalFS.pdf, UPDATELOG.md (docs)
• pyCausalFS/ (main package)
• pyBN/ (standalone Bayes‐net toolkit subpackage)
Under pyCausalFS/:
• CBD/ – Constraint‐based MB & PC learning (modules: MBs, data, example, output, evaluation scripts)
• GSL/ – Global‐structure learning (GSBN, MMHC, etc.)
• LSL/ – Local‐structure learning (MB‐by‐MB, PCD‐by‐PCD, CMB)
• SSD/ – Score‐based MB & PC learning (S2TMB, SLL, plus embedded pyBN usage)
Common patterns:
• Each algorithm family has subfolders for its implementations (MBs or PC methods), a common sub‐package for statistics/tests, a data folder for sample graphs & CSV, and example/output scripts.
• pyBN is a full BN engine (classes, io, learning, inference, classification, plotting) that SSD, GSL, and LSL modules depend on for some routines.
• No frontend/backend or service separation—this is a library to be imported and called by user scripts.
README for additional insights
• The README outlines 7 algorithm categories (constraint‐based MB, multiple MB, constraint‐based PC, score‐based MB/PC, local BN, local‐to‐global BN).
• It calls out manual documentation for low‐level details.
• Core dependencies (e.g., NumPy, SciPy, pandas) are implied but not shown—assume these as external services.
Elements
• Algorithm modules (e.g., blue)
• Common utilities (green)
• Data & I/O (orange)
• External/dependency modules (gray)