CDANs: Temporal Causal Discovery from Autocorrelated and Non-Stationary Time Series Data

A small patch release. Quality-of-life improvements with no algorithm or API changes: a runnable CSV workflow example, a cleaner homepage demo, fixes to documentation that misrepresented the implementation status, and dynamic PyPI badges in the README. Added CSV workflow. New runnable example examples/from_csv.py and a docs page Using your own CSV data showing how to load a CSV with pandas, pass column names through to CDANs, and read out the recovered graph with named edges instead of integer indices. Dynamic PyPI badges in README. Five badges (version, supported Python versions, license, monthly downloads, development status), all pulled live from PyPI metadata so they auto-update on each release. Fixed Documentation falsely claimed infer_nonsta_dir.m was not ported. The independent-change-principle direction inference is in fact ported in full, including both the fixed-bandwidth path and (since v0.1.0) the GP-bandwidth path via scikit-learn. Fixed the implementation-status table in README and the corresponding sections of docs/index.md, docs/experiments.md, and docs/api/independent_change.md. Broken Changelog URL on PyPI. pyproject.toml previously pointed at /blob/main/CHANGELOG.md but the default branch is master. The Changelog link in the PyPI sidebar now resolves correctly. Homepage evaluation demo showed weak results. The previous demo used an n_vars=5, n_samples=400, tau_max=2 configuration that sits on the edge of KCI's detection limits and gave partial recovery (0 of 2 contemporaneous edges, 50 percent changing-modules FDR). Replaced with n_vars=4, n_samples=800, tau_max=1, n_changing=1, seed=7, which gives perfect recovery in roughly 25 seconds. Realistic 6-variable benchmark numbers remain on the experiments page. Install pip install --upgrade cdans Full changelog https://github.com/hferdous/CDANs/compare/v0.1.0...v0.1.1