magenpy 0.2.3 on Python PyPI

[0.2.3] - 2026-10-27

Added parsers for REGENIE step-2 association results and LDAK summary
statistics. The LDAK parser supports SumHer/MegaPRS files, full .assoc
output, and all four documented statistic layouts (Z, BETA/SE,
Stat/Direction, and P/Direction).
Added a central summary-statistics parser registry with dynamic format and
alias discovery through SumstatsTable and GWADataLoader. New formats no
longer require duplicated supported-format lists in those classes.
Added a native C++/Cython identity-indexer check for contiguous 32-bit and
64-bit integer arrays, with a bounded-memory Python fallback for other array
representations.

Expanded parser dtype declarations across summary statistics, annotations,
variant filters, sample filters, LD blocks, and cluster assignments. Known
numeric columns now avoid pandas dtype inference, identifiers retain leading
zeroes, positions use compact 32-bit representations, and caller-provided
dtype overrides remain authoritative.
Reduced harmonization allocations by checking identity indexers in one native
pass instead of constructing full arange arrays, retaining an existing
summary-statistics dataframe when its ordering is already correct, and
updating signed statistics and allele frequencies only at flipped positions.
Improved parser defaults and diagnostics for tabular inputs while preserving
whitespace-delimited defaults for native REGENIE and LDAK files.

Preserved textual SNP and sample identifiers when reading filter and cluster
files, preventing numeric coercion from changing identifiers with leading
zeroes.
Correctly converted REGENIE LOG10P values to p-values and LDAK .assoc
mean A1 allele counts to effect-allele frequencies.