pirl-unc / hitlist / 26992506673
80%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 29
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 76.645% (-0.2%) from 76.832%
26992506673

push

github

web-flow 
v1.30.62: int-encoded columnar k-mer index (#250, foundation for #248) (#273)

Replace ProteomeIndex's dict[str, int|np.ndarray] k-mer index with a
per-length columnar structure (_PackedIndex): each k-mer packs into a
single uint64 code (5 bits/residue), stored as code-sorted uint64 arrays
+ int64 offsets + int64 packed-posting values.

- Build is vectorized (numpy rolling-pack per protein, ~1.8s for 20M
  residues) + one stable sort per length with a boundary-scan grouping
  that avoids np.unique's redundant re-sort. Cold build ~1.2-1.4x faster;
  the grouping is fundamentally sort-bound (173M k-mer occurrences), so
  the bigger payoff is the warm-load format change in #248.
- Query is np.searchsorted on the sorted codes (no Python dict).
- Encoding verified collision-free for the real alphabet (21 residues,
  k<=11 -> <=55 bits). Legacy string-dict build retained as a fallback
  for inputs outside the budget (alphabet>32 or longer lengths) or with
  non-standard residues.

Public API unchanged: lookup/map_peptides/all_kmers/merge/__len__/
__contains__ all preserved; pickle round-trips. Verified: 54
test_proteome.py contract tests pass, full suite 866 passed, and packed
vs legacy lookup parity is exact (48,129,615 k-mers, ALL MATCH over 5003
peptides on the Equus caballus proteome).

Stage 1 of the proteome-index redesign (see tasks/). #248 swaps the disk
cache to mmap'd Arrow next; #176/#63 re-profile after.

4821 of 6290 relevant lines covered (76.65%)

0.77 hits per line

Coverage Regressions

Lines Coverage ∆ File
83
83.8
 -5.34% proteome.py
Jobs
ID Job ID Ran Files Coverage
1 26992506673.1 29
76.65
 GitHub Action Run
Source Files on build 26992506673