openvax / topiary / 24419453085

88%

master: 90%

push

github

CachedPredictor — predictions-as-data (#130) — v5.5.0

Part 1 of #128 (pluggable prediction sources).  Core CachedPredictor
class + four loaders + documentation.

Core invariant: a single CachedPredictor holds predictions from
exactly one (prediction_method_name, predictor_version) pair; None /
NaN / empty-string values rejected.  Explicit opt-in equivalence via
also_accept_versions for rc→final and model-data reflashes.

Loaders: from_dataframe, from_topiary_output (Parquet/TSV), from_tsv
(generic w/ column mapping), from_mhcflurry (auto-composes
predictor_version from the installed mhcflurry + active model
release via topiary.mhcflurry_composite_version()).

Fallback: pass fallback=<predictor> to delegate cache misses; results
merged back so subsequent queries serve locally.  Pure read-through
mode (empty cache + fallback) is supported; identity is discovered
from the fallback's first output.

Filed openvax/mhctools#193 for the predict_peptides_dataframe schema
asymmetry the CachedPredictor backfills internally.

Deferred to follow-up PRs: NetMHCpan/NetMHC/NetMHCstabpan/
NetMHCIIpan parsers, sharding (concat / from_directory).

1090 tests passing (up from 1052), lint clean, mkdocs --strict clean.

209 of 230 new or added lines in 2 files covered. (90.87%)

3214 of 3640 relevant lines covered (88.3%)

2.65 hits per line