built rag search setup

2026-06-12 02:45:44 -04:00
parent e8f5a3d334
commit 7010a4f3b9
2 changed files with 407 additions and 0 deletions
@@ -0,0 +1,371 @@
+"""Hybrid search orchestration."""
+
+from __future__ import annotations
+
+import logging
+import re
+from concurrent.futures import ThreadPoolExecutor
+from dataclasses import dataclass, replace
+from typing import TYPE_CHECKING
+
+from pgvector.sqlalchemy import Vector
+from sqlalchemy import literal, select
+from sqlalchemy.orm import Session
+
+from python.ebook_search.bm25_corpus import (
+    load_bm25_corpus,
+    score_bm25_corpus,
+)
+from python.ebook_search.embeddings import MODEL_DIMENSIONS, embed_query, get_embedding_table
+from python.ebook_search.rerank import rerank_chunks
+from python.ebook_search.timing import RuntimeStep, timed_result
+from python.orm.richie import (
+    EbookChapter,
+    EbookChunk,
+    EbookEmbeddingModel,
+    EbookSource,
+)
+
+if TYPE_CHECKING:
+    from collections.abc import Mapping
+
+    from sqlalchemy.engine import Engine
+
+    from python.ebook_search.config import EbookSearchConfig
+
+logger = logging.getLogger(__name__)
+BM25_CANDIDATE_LIMIT = 120
+
+
+@dataclass(frozen=True)
+class SearchResult:
+    """One source chunk returned by search."""
+
+    chunk_id: int
+    text: str
+    source_title: str
+    score: float = 0.0
+    vector_score: float | None = None
+    bm25_score: float | None = None
+    fused_score: float | None = None
+    rerank_score: float | None = None
+    source_author: str | None = None
+    chapter_title: str | None = None
+    page_label: str | None = None
+    rank_source: str = "Hybrid"
+
+
+@dataclass(frozen=True)
+class SearchResponse:
+    """Search output for the UI."""
+
+    query: str
+    results: list[SearchResult]
+    rank_label: str
+    timings: tuple[RuntimeStep, ...] = ()
+
+    @property
+    def total_runtime_ms(self) -> float:
+        """Return total measured runtime for the response."""
+        return sum(step.duration_ms for step in self.timings if step.counts_toward_total)
+
+
+@dataclass(frozen=True)
+class RetrievalResponse:
+    """Parallel retrieval output for vector and BM25 candidates."""
+
+    vector_results: list[SearchResult]
+    lexical_results: list[SearchResult]
+    timings: tuple[RuntimeStep, ...]
+
+
+def search_ebooks(
+    engine: Engine,
+    query: str,
+    config: EbookSearchConfig,
+    *,
+    rerank: bool = False,
+) -> SearchResponse:
+    """Run hybrid vector/BM25 search and optional reranking."""
+    if not query.strip():
+        logger.info("ebook_search_empty_query")
+        return SearchResponse(query=query, results=[], rank_label="Hybrid")
+
+    logger.info("ebook_search_start query_length=%s rerank=%s", len(query), rerank)
+    timings: list[RuntimeStep] = []
+    retrieval_query, timing = timed_result("Query preparation", retrieval_query_from_text, query)
+    timings.append(timing)
+    retrieval, timing = timed_result(
+        "Hybrid retrieval",
+        parallel_retrieval,
+        engine,
+        retrieval_query,
+        config,
+    )
+    timings.extend(retrieval.timings)
+    timings.append(timing)
+    fused, timing = timed_result(
+        "Reciprocal rank fusion",
+        reciprocal_rank_fusion,
+        retrieval.vector_results,
+        retrieval.lexical_results,
+    )
+    timings.append(timing)
+    if config.rerank.enabled and rerank:
+        response, timing = timed_result("Rerank", apply_rerank, query, fused, config)
+    else:
+        response, timing = timed_result("Rerank skipped", skip_rerank, query, fused, config)
+    timings.append(timing)
+    response = replace(response, timings=tuple(timings))
+    logger.info(
+        "ebook_search_complete vector_candidates=%s lexical_candidates=%s "
+        "fused_candidates=%s returned=%s rank_label=%s runtime_ms=%.1f",
+        len(retrieval.vector_results),
+        len(retrieval.lexical_results),
+        len(fused),
+        len(response.results),
+        response.rank_label,
+        response.total_runtime_ms,
+    )
+    return response
+
+
+def parallel_retrieval(engine: Engine, query: str, config: EbookSearchConfig) -> RetrievalResponse:
+    """Run vector and BM25 candidate retrieval concurrently with separate database sessions."""
+    with ThreadPoolExecutor(max_workers=2, thread_name_prefix="ebook-search") as executor:
+        vector_future = executor.submit(
+            timed_result,
+            "Embedding + vector search",
+            vector_candidates,
+            engine,
+            query,
+            config,
+        )
+        bm25_future = executor.submit(
+            timed_result,
+            "BM25 search",
+            bm25_candidates,
+            query,
+            config,
+        )
+        vector_results, vector_timing = vector_future.result()
+        lexical_results, lexical_timing = bm25_future.result()
+
+    logger.info(
+        "ebook_parallel_retrieval_complete vector_candidates=%s lexical_candidates=%s",
+        len(vector_results),
+        len(lexical_results),
+    )
+    return RetrievalResponse(
+        vector_results=vector_results,
+        lexical_results=lexical_results,
+        timings=(
+            replace(vector_timing, counts_toward_total=False),
+            replace(lexical_timing, counts_toward_total=False),
+        ),
+    )
+
+
+def skip_rerank(
+    query: str,
+    candidates: list[SearchResult],
+    config: EbookSearchConfig,
+) -> SearchResponse:
+    """Return fused hybrid results without reranking."""
+    logger.info("ebook_rerank_skipped candidates=%s", len(candidates))
+    return SearchResponse(query=query, results=candidates[: config.top_k], rank_label="Hybrid")
+
+
+def apply_rerank(
+    query: str,
+    candidates: list[SearchResult],
+    config: EbookSearchConfig,
+) -> SearchResponse:
+    """Rerank already-fused hybrid candidates."""
+    reranked = rerank_chunks(query, candidates[: config.rerank.candidates], config.rerank)
+    logger.info(
+        "ebook_rerank_complete input_candidates=%s returned=%s",
+        min(len(candidates), config.rerank.candidates),
+        len(reranked),
+    )
+    return SearchResponse(
+        query=query,
+        results=[replace(result, rank_source="Hybrid + rerank") for result in reranked[: config.top_k]],
+        rank_label="Hybrid + rerank",
+    )
+
+
+def vector_candidates(engine: Engine, query: str, config: EbookSearchConfig) -> list[SearchResult]:
+    """Return pgvector cosine candidates for a normalized query."""
+    with Session(engine) as session:
+        model = session.scalar(select(EbookEmbeddingModel).where(EbookEmbeddingModel.name == config.embedding_model))
+        if model is None:
+            msg = f"Embedding model is not registered: {config.embedding_model}"
+            raise ValueError(msg)
+
+        expected_dimension = MODEL_DIMENSIONS[config.embedding_model]
+        if model.dimension != expected_dimension:
+            msg = f"Model row dimension {model.dimension} does not match configured dimension {expected_dimension}"
+            raise ValueError(msg)
+
+        embedding = embed_query(query, config)
+        limit = max(config.rerank.candidates, config.top_k) * 4
+        embedding_table = get_embedding_table(model.dimension)
+
+        embedding_param = literal(embedding, type_=Vector(model.dimension))
+        distance = embedding_table.embedding.op("<=>")(embedding_param)
+        score = (literal(1.0) - distance).label("score")
+        statement = (
+            select(
+                EbookChunk.id.label("chunk_id"),
+                EbookChunk.text.label("text"),
+                EbookSource.title.label("source_title"),
+                EbookSource.author.label("source_author"),
+                EbookChapter.title.label("chapter_title"),
+                EbookChunk.page_label.label("page_label"),
+                score,
+            )
+            .select_from(embedding_table)
+            .join(EbookChunk, EbookChunk.id == embedding_table.chunk_id)
+            .join(EbookSource, EbookSource.id == EbookChunk.source_id)
+            .outerjoin(EbookChapter, EbookChapter.id == EbookChunk.chapter_id)
+            .where(embedding_table.model_id == model.id)
+            .order_by(distance)
+            .limit(limit)
+        )
+        rows = session.execute(statement).mappings()
+        results = [search_result_from_row(row) for row in rows]
+        logger.info(
+            "ebook_vector_search_complete model=%s dimension=%s candidates=%s",
+            config.embedding_model,
+            model.dimension,
+            len(results),
+        )
+        return results
+
+
+def bm25_candidates(query: str, config: EbookSearchConfig) -> list[SearchResult]:
+    """Return BM25-ranked lexical candidates using the persisted corpus."""
+    corpus = load_bm25_corpus(config)
+    if not corpus.records:
+        logger.info("ebook_bm25_search_complete corpus=0 candidates=0")
+        return []
+
+    scored_records = score_bm25_corpus(query, corpus, limit=BM25_CANDIDATE_LIMIT)
+    results = [
+        replace(search_result_from_row(record), score=score, vector_score=None, bm25_score=score)
+        for record, score in scored_records
+    ]
+
+    max_score = results[0].bm25_score if results else 0.0
+    logger.info(
+        "ebook_bm25_search_complete corpus=%s candidates=%s max_score=%.6f",
+        len(corpus.records),
+        len(results),
+        max_score,
+    )
+    return results
+
+
+def reciprocal_rank_fusion(
+    vector_results: list[SearchResult],
+    lexical_results: list[SearchResult],
+    *,
+    rank_constant: int = 60,
+) -> list[SearchResult]:
+    """Fuse vector and lexical rankings with Reciprocal Rank Fusion."""
+    by_chunk: dict[int, SearchResult] = {}
+    scores: dict[int, float] = {}
+    vector_scores: dict[int, float] = {}
+    bm25_scores: dict[int, float] = {}
+
+    for rank, result in enumerate(vector_results, start=1):
+        by_chunk.setdefault(result.chunk_id, result)
+        vector_scores[result.chunk_id] = result.vector_score if result.vector_score is not None else result.score
+        scores[result.chunk_id] = scores.get(result.chunk_id, 0.0) + (1 / (rank_constant + rank))
+
+    for rank, result in enumerate(lexical_results, start=1):
+        by_chunk.setdefault(result.chunk_id, result)
+        bm25_scores[result.chunk_id] = result.bm25_score if result.bm25_score is not None else result.score
+        scores[result.chunk_id] = scores.get(result.chunk_id, 0.0) + (1 / (rank_constant + rank))
+
+    return sorted(
+        (
+            replace(
+                result,
+                score=scores[result.chunk_id],
+                vector_score=vector_scores.get(result.chunk_id),
+                bm25_score=bm25_scores.get(result.chunk_id),
+                fused_score=scores[result.chunk_id],
+                rank_source="Hybrid",
+            )
+            for result in by_chunk.values()
+        ),
+        key=lambda result: result.score,
+        reverse=True,
+    )
+
+
+def search_result_from_row(row: Mapping[str, object]) -> SearchResult:
+    """Convert a database row mapping into a search result."""
+    return SearchResult(
+        chunk_id=int(row["chunk_id"]),
+        text=str(row["text"]),
+        source_title=str(row["source_title"]),
+        source_author=optional_str(row["source_author"]),
+        chapter_title=optional_str(row["chapter_title"]),
+        page_label=optional_str(row["page_label"]),
+        score=float(row["score"]) if "score" in row else 0.0,
+        vector_score=float(row["score"]) if "score" in row else None,
+    )
+
+
+def optional_str(value: object) -> str | None:
+    """Convert nullable database values to optional strings."""
+    if value is None:
+        return None
+    return str(value)
+
+
+TOKEN_RE = re.compile(r"[A-Za-z0-9_]+")
+
+
+def tokens(text_value: str) -> list[str]:
+    """Extract tokens from a text value.
+
+    This is a simple approximation of the tokenization used by PostgreSQL's full-text search,
+    which is sufficient for BM25 candidate retrieval. It lowercases tokens and includes alphanumeric characters and
+    underscores.
+    """
+    return [match.group(0).lower() for match in TOKEN_RE.finditer(text_value)]
+
+
+QUERY_STOP_WORDS = {
+    "a",
+    "an",
+    "and",
+    "are",
+    "as",
+    "at",
+    "does",
+    "for",
+    "in",
+    "is",
+    "of",
+    "the",
+    "to",
+    "what",
+    "when",
+    "where",
+    "which",
+    "who",
+    "why",
+}
+
+
+def retrieval_query_from_text(query: str) -> str:
+    """Remove generic question words while preserving entity and series terms."""
+    keywords = [token for token in tokens(query) if token not in QUERY_STOP_WORDS]
+    if not keywords:
+        return query
+    return " ".join(keywords)
@@ -0,0 +1,36 @@
+"""Runtime timing helpers for EPUB search."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from time import perf_counter
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from collections.abc import Callable
+
+
+@dataclass(frozen=True)
+class RuntimeStep:
+    """Elapsed runtime for one named search step."""
+
+    name: str
+    duration_ms: float
+    counts_toward_total: bool = True
+
+
+def runtime_step_from_start(name: str, start_seconds: float) -> RuntimeStep:
+    """Create a runtime step from a prior perf_counter timestamp."""
+    return RuntimeStep(name=name, duration_ms=(perf_counter() - start_seconds) * 1000)
+
+
+def timed_result[T, **P](
+    name: str,
+    operation: Callable[P, T],
+    *args: P.args,
+    **kwargs: P.kwargs,
+) -> tuple[T, RuntimeStep]:
+    """Run an operation and return its result plus elapsed runtime."""
+    start_seconds = perf_counter()
+    result = operation(*args, **kwargs)
+    return result, runtime_step_from_start(name, start_seconds)