openvax / topiary
91%
master: 93%

LAST BUILD BRANCH: v5.59.2
DEFAULT BRANCH: master
Repo Added
Files 43
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

LAST BUILD ON BRANCH context-forwarding
branch: context-forwarding
CHANGE BRANCH
x

coverage: 91.023% (+0.1%) from 90.897%
33406597978

Pull #212

github

iskandr 
Share one EvalContext across operations; forward default_methods (#179, #178)

#179: apply_filter/apply_sort/evaluate_scores each built their own
EvalContext, recomputing the grouping every call. They now accept
context=, and EvalContext.derive() lets a filter (which needs
filter_context=True) and a sort (which needs it False) share one
grouping without either flipping the other's flag.

The reuse is narrower than #179's example implies, and the code says so
rather than letting a caller find out: apply_filter and apply_sort both
return new frames, so a context cannot be threaded down a
filter -> sort -> score pipeline. Passing a stale one raises on an
identity check instead of mapping rows to another frame's groups.

#178: a predictor running two models that produce the same kind made
every unqualified reference in sort_by ambiguous with no way to resolve
it through the predictor. TopiaryPredictor takes default_methods, and
TopiaryResult.filter_by/sort_by take default_methods and group_keys.
No group_keys on the predictor: it builds its own frame, so the
inferred grouping is right by construction.

27 tests across tests/test_shared_context.py and
tests/test_context_forwarding.py. Version 5.29.0.

Claude-Session: https://claude.ai/code/session_01BAUqjqDDPXaTdtUDJqapJG
Pull Request #212: Share one EvalContext across operations; forward default_methods (#179, #178)

37 of 37 new or added lines in 5 files covered. (100.0%)

5597 of 6149 relevant lines covered (91.02%)

2.73 hits per line

Relevant lines Covered
6149 RELEVANT LINES 5597 COVERED LINES
2.73 HITS PER LINE
Source Files on context-forwarding

Recent builds

Builds Branch Commit Type Ran Committer Via Coverage
33406597978 context-forwarding Share one EvalContext across operations; forward default_methods (#179, #178) #179: apply_filter/apply_sort/evaluate_scores each built their own EvalContext, recomputing the grouping every call. They now accept context=, and EvalContext.derive() ... Pull #212 iskandr github
91.02
See All Builds (561)