openvax / topiary / 24170771911

85%

master: 90%

Build Type

Pull #97

github

Committed by iskandr

Commit Message

Add multi-model disambiguation for ranking/filtering DSL

When multiple MHC binding predictors produce the same prediction kind
(e.g. both NetMHCpan and MHCflurry produce pMHC_affinity), the ranking
system now supports qualifying expressions by prediction method.

Python API — bracket syntax on KindAccessor:

    Affinity["netmhcpan"] <= 500
    Affinity["mhcflurry"].score
    Affinity["netmhcpan"].norm(500, 200)
    0.5 * Affinity["netmhcpan"] + 0.5 * Affinity["mhcflurry"]

CLI — tool_kind underscore syntax:

    --ranking "netmhcpan_affinity <= 500 | mhcflurry_el.rank <= 2"
    --rank-by "netmhcpan_affinity,mhcflurry_presentation"

All existing short aliases work with tool prefix: netmhcpan_ba,
mhcflurry_el, netmhcpan_aff, etc.

Error handling:
- Unqualified access with multiple models producing the same kind
  raises ValueError listing available methods
- Qualified access with a typo raises ValueError with "Did you mean"
  suggestions via difflib fuzzy matching
- Unqualified access with a single model works without qualification
  (backwards compatible)

Adds 32 new tests including adversarial edge cases for ambiguity
detection, typo suggestions, empty data, missing columns, cross-model
composites, and case-insensitive matching.

Pull Request Pull Request #97: Multi-model disambiguation for ranking/filtering DSL

Coverage Stats

95 of 117 new or added lines in 2 files covered. (81.2%)

5 existing lines in 1 file now uncovered.

1162 of 1360 relevant lines covered (85.44%)

2.56 hits per line