jzombie / rust-triplets / 29175230798

96%

Build Type

push

github

Committed by web-flow

Commit Message

Add HuggingFace dict dataset support, weighted scheduling, and explicit dot-path columns (#124)

* add HuggingFace dict dataset support, weighted scheduling, and explicit
dot-path columns

- Add `negative=` column mapping for HF dict datasets (e.g.
  QQP_triplets) where
  negatives are embedded as list values in the same row; list elements
  are expanded
  into individual Context sections
- Add `NegativeStrategy::SameRecord` which picks negative chunks from
  the anchor
  record's own Context sections
- Add `weight=` URI key for per-source weighted scheduling;
  `build_hf_sources_with_weights`
  returns both sources and a weight map for use with
  `advance_with_weights`
- Change `resolve_json_path` from auto-scan to explicit dot-path
  resolution:
  `set.query` walks `row["set"]["query"]`, bare `query` does top-level
  only;
  no implicit nested dict scanning
- Update README with dict dataset examples, negative= key docs, weight=
  key docs,
  SameRecord strategy, user-derived negatives section, and weighted
  scheduling example
- Fix rustdoc broken intra-doc links
  (`TripletSampler::advance_with_weights`)
- Security: bump crossbeam-epoch 0.9.18→0.9.20, quinn-proto
  0.11.14→0.11.16;
  add .cargo/audit.toml and .cargo/deny.toml for transitive dep audit
  config

* Fix incorrect API references

Coverage Stats

339 of 354 new or added lines in 2 files covered. (95.76%)

9 existing lines in 1 file now uncovered.

20812 of 21734 relevant lines covered (95.76%)

5309.12 hits per line