Conversation
…on dataset MLP reranker (128-64-32 MLP, isotonic calibration) replaces LLM verification for borderline candidates. Trained on 3,600 labeled query-document pairs across 4 legal domains. Achieves MRR 0.933 vs 0.665 for semantic-only search (+40%) at zero inference cost — proving learned retrieval outperforms LLM reranking. New files: - rag_dependencies/feature_extractor.py: 15-feature vector extraction (75 tests) - benchmarks/eval_dataset.json: 180 labeled queries across 4 domains - benchmarks/eval_dataset_schema.py: Pydantic validation for dataset - benchmarks/generate_eval_dataset.py: Dataset validate/expand/stats tool - benchmarks/train_reranker.py: Training pipeline (3 models, cross-val, calibration) - benchmarks/run_baseline.py: Baseline measurement (P@k, MRR, latency, cost) - benchmarks/run_ablation_full.py: 7-strategy ablation study - benchmarks/cost_comparison.py: Before/after cost analysis - benchmarks/retrain_monthly.py: Automated monthly retraining from MongoDB - benchmarks/generate_graphs.py: Cost/latency/MRR comparison graphs - media/cost_latency_comparison.png: Visual comparison (ARF vs MongoDB) - models/.gitkeep: Model directory Modified files: - rag_dependencies/query_processor.py: MLP integration in _apply_main_abc_gates - config.py: MLP config keys (use_mlp_reranker, uncertainty thresholds) - config_schema.py: Optional MLP Pydantic fields - README.md: Summary, ablation results, MLP architecture, graphs Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…ency - query_processor.py: Wire MLP into _apply_main_abc_gates with graceful fallback - config.py: Add use_mlp_reranker, mlp_uncertainty_low/high to all 4 domains - config_schema.py: Add optional MLP Pydantic fields - train_reranker.py: Use FeatureExtractor (15-dim) instead of hand-crafted 22-dim features, fix eval_dataset.json format support, add prefix title matching - README.md: Add summary, ablation table, MLP architecture, comparison graphs Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.