pdex-0.2.5

What's Changed

• Add CPM floor filter (cpm_filter) by @LeonHafner in #77

Full Changelog: pdex-0.2.4...pdex-0.2.5

pdex-0.2.4

What's Changed

• Fix numba thread crash on hosts with restricted CPU affinity by @LeonHafner in #76

Full Changelog: pdex-0.2.3...pdex-0.2.4

pdex-0.2.3

What's Changed

• Define 0/0 fold change and percent change as 0.0 instead of NaN by @LeonHafner in #75

Full Changelog: pdex-0.2.2...pdex-0.2.3

pdex-0.2.2

Important: fold_change column is deprecated

The fold_change column has held log2(target/ref) since 0.2.0 despite the misleading name. This silently broke downstream consumers that assumed it was a linear ratio (notably cell-eval ≤ 0.7.0, which double-logged the values and zeroed out roughly half of all DE results — see ArcInstitute/cell-eval#232 for the full discussion).

This release adds an explicit log2_fold_change column with the same values, and emits a FutureWarning on every pdex(...) call.

Migration

pdex version	fold_change	log2_fold_change
< 0.2.0	linear ratio	(not present)
0.2.0 – 0.2.1	log2 (misnamed)	(not present)
0.2.2 (this release)	log2, deprecated alias	log2
0.3.0 (next minor)	removed	log2

Downstream code should switch to log2_fold_change now:

# old (still works in 0.2.2, removed in 0.3.0)
df["fold_change"]

# new
df["log2_fold_change"]

What's Changed

• Rename fold_change column to log2_fold_change (deprecate alias) by @LeonHafner in #74

New Contributors

• @LeonHafner made their first contribution in #74

Full Changelog: pdex-0.2.1...pdex-0.2.2

pdex-0.2.1

What's Changed

• feat: add epsilon parameter to dampen sparsity-artifact fold changes by @sullivanj91 in #72
• ci: enforce semver bump on pull requests by @sullivanj91 in #73

New Contributors

• @sullivanj91 made their first contribution in #72

Full Changelog: pdex-0.2.0...pdex-0.2.1

pdex-0.2.0

What's Changed

• Integrate pdex2 by @noamteyssier in #68
  • numba backend is stabilized and is the only path through pdex now (implementation in https://github.com/ArcInstitute/numba-mwu)
  • only statistical test supported is MWU (other approaches were not maintained and likely error prone)
  • Sparse matrix operations are prioritized and fully supported with CSR
  • API has breaking changes but much reduced configuration to improve clarity
  • Polars output is defaulted over pandas
  • Introduced new modes ("ref", "all", "on_target") to change differential expression scheme
  • Improved is_log1p test with warning that is quieted on explicit configuration
• Feat/backed adata by @noamteyssier in #69
  • Supported in new version and reduce amount of double instantiations (when using "ref" mode)
• Pdex 0.2.0 by @noamteyssier in #70

Full Changelog: pdex-0.1.28...pdex-0.2.0

pdex-0.1.28

What's Changed

• Add Backed AnnData Support via Low-Memory Chunked Processing and Numba Kernels by @nick-youngblut in #64

New Contributors

• @nick-youngblut made their first contribution in #64

Full Changelog: pdex-0.1.27...pdex-0.1.28

pdex-0.1.27

What's Changed

• 60 fix invalid p value by @noamteyssier in #61

Full Changelog: pdex-0.1.26...pdex-0.1.27

pdex-0.1.26

What's Changed

• Update _single_cell.py by @MingzeDong in #58

New Contributors

• @MingzeDong made their first contribution in #58

Full Changelog: pdex-0.1.25...pdex-0.1.26

pdex-0.1.25

What's Changed

• 56 fix log transformation guess by @noamteyssier in #57

Full Changelog: pdex-0.1.24...pdex-0.1.25