openvax / topiary / 24905785512

90%

push

github

EvalContext default_methods for multi-predictor frames (#140) (#141)

* EvalContext default_methods for multi-predictor frames (#140) — v5.10.0

- EvalContext(df, default_methods={...}) — per-kind default
  prediction_method_name resolves unqualified Field references when a
  DataFrame has multiple methods per kind. Without it, ambiguous
  references still raise (behavior unchanged).
- Keys accept canonical kind names ("pMHC_affinity"), DSL short names
  ("affinity", "ba", "el", ...), or mhctools Kind constants.
- apply_filter / apply_sort forward the kwarg.
- Ambiguity error message now points users at default_methods.
- CHANGELOG + version bump to 5.10.0.

* Expand default_methods synonym coverage + add 'processing' alias

Parameterized tests pin every accepted spelling of each kind:

- pMHC_affinity: 'pMHC_affinity', 'pmhc_affinity', 'affinity',
  'ba', 'aff', 'ic50', Kind.pMHC_affinity
- pMHC_presentation: 'pMHC_presentation', 'presentation', 'el'
- pMHC_stability: 'pMHC_stability', 'stability'
- antigen_processing: 'antigen_processing', 'processing'

'processing' was the one short name that _build_kind_aliases didn't
emit (the '_short_name' helper strips 'pmhc_' but leaves
'antigen_processing' untouched). Added it explicitly alongside 'el',
'ba', 'aff', 'ic50'.

* Address PR #141 review: explicit regex=False + short aliases in error

1. Pin regex=False on both str.contains calls in Field.eval so the
   substring match for prediction_method_name is literal, not a regex.
   Previously 'net.*pan' would have regex-matched 'netmhcpan' silently;
   now it raises method-not-found as users would expect.

2. default_methods unknown-kind error now lists every accepted
   spelling (short aliases + canonical kind names), so 'banana' ->
   '[aff, affinity, antigen_processing, ba, el, ..., pMHC_affinity,
   ..., processing, stability, ...]' rather than just the canonical
   subset. Lower-case duplicates of canonicals are filtered out for
   readability.

36 of 36 new or added lines in 3 files covered. (100.0%)

3713 of 4197 relevant lines covered (88.47%)

2.65 hits per line