Skip to content

Releases: vignesh2027/datamend.py

v1.1.3

01 Jun 10:36

Choose a tag to compare

Fixed

  • test_plugins.py select_dtypes "str" TypeError — plugin test used
    include=["object", "str"]; changed to "string" for pandas 2.x/3.x compat.
  • All-null object column fillna(None) crash_NullDetector.fix() computed
    series.mode() on all-null columns which returns [None] in pandas 2.x; calling
    fillna(None) raised ValueError: Must specify a fill 'value' or 'method'.
    Fixed by using series.dropna().mode() and skipping imputation when no non-null
    reference values exist.

v1.1.2

01 Jun 10:19

Choose a tag to compare

Fixed

  • Cross-platform select_dtypes TypeErrorrepair.py used include=["object", "str"]
    where "str" maps to numpy.str_ (rejected) in pandas 2.x. Changed to "string"
    (pandas StringDtype) which works in both pandas 2.x and 3.x across Python 3.9–3.13
    on Windows, macOS, and Linux.
  • Windows UTF-8 HTML comparison testtest_stress.py read HTML files without
    encoding="utf-8", causing · and to appear as mojibake on Windows (cp1252).
    Now reads with explicit encoding="utf-8".
  • Pipeline overall_mend_score float overflow — clamped to [0.0, 100.0] via
    min(100.0, max(0.0, raw)) to prevent 100.00000000000001 assertion failures.
  • 131 new advanced tests added for all four pillars, CLI, cross-pillar integration,
    serialisation round-trips, adversarial inputs (emoji, unicode, all-null, constant cols,
    single row/col, 100k rows, boolean/datetime/wide DataFrames).

v1.1.1

15 May 15:40

Choose a tag to compare

Fixed

  • Ruff lint compliance — removed all ANN annotation rules from ruff config and
    ran auto-fix + unsafe-fixes across all modules (integrations/, pipeline.py,
    plugins/, report.py, core/__init__.py, core/contract.py, core/drift.py).
    All 386 lint errors resolved; ruff check datamend/ now exits clean.
  • DriftRadar.detect single-arg form — when .fit(reference_df) has been called,
    detect(prod_df) now correctly uses the stored reference without requiring two args.
  • GitHub Actions CI workflow — removed ANN select rules from ruff; CI lint step
    now passes on Python 3.9–3.13 across Ubuntu, macOS, and Windows.
  • GitHub Actions publish workflow — replaced OIDC trusted publishing with
    twine upload via PYPI_API_TOKEN secret; added artifact upload/download between
    the publish and github-release jobs so dist/ is available in both.