Add BM25SRetriever: pure-Python BM25 with no Java/Pyserini dependency

pyproject.toml

@@ -80,6 +80,9 @@ retriever = [
   # Sparse retrieval (BM25)
   "pyserini==0.43.0",
 
+  # Sparse retrieval (BM25S - pure Python, no Java dependency)
+  "bm25s>=0.2.0",
+
   # Dense retrieval 
   "faiss-cpu==1.9.0.post1",
   "h5py==3.12.1",

rankify/retrievers/__init__.py

@@ -1,28 +1,82 @@
-# rankify/retrievers/__init__.py - MODIFIED VERSION
+# rankify/retrievers/__init__.py
 
 from .retriever import Retriever
 from .base_retriever import BaseRetriever
-from .bm25_retriever import BM25Retriever
-from .dense_retriever import DenseRetriever
-from .ance_retriever import ANCERetriever  # NEW IMPORT
-from .bge_retriever import BGERetriever
-from .colbert_retriever import ColBERTRetriever
-from .contriever_retriever import ContrieverRetriever
-from .online_retriever import OnlineRetriever
-from .hyde_retriever import HydeRetriever
-from .diver_dense_retriever import DiverDenseRetriever
-from .diver_bm25_retriever import DiverBM25Retriever
-from .reasonir_retriever import ReasonIRRetriever
-from .reasonembed_retriever import ReasonEmbedRetriever
-from .bge_reasoner_retriever import BgeReasonerRetriever
+from .bm25s_retriever import BM25SRetriever
+
+try:
+    from .bm25_retriever import BM25Retriever
+except ImportError:
+    BM25Retriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .dense_retriever import DenseRetriever
+except ImportError:
+    DenseRetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .ance_retriever import ANCERetriever
+except ImportError:
+    ANCERetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .bge_retriever import BGERetriever
+except ImportError:
+    BGERetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .colbert_retriever import ColBERTRetriever
+except ImportError:
+    ColBERTRetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .contriever_retriever import ContrieverRetriever
+except ImportError:
+    ContrieverRetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .online_retriever import OnlineRetriever
+except ImportError:
+    OnlineRetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .hyde_retriever import HydeRetriever
+except ImportError:
+    HydeRetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .diver_dense_retriever import DiverDenseRetriever
+except ImportError:
+    DiverDenseRetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .diver_bm25_retriever import DiverBM25Retriever
+except ImportError:
+    DiverBM25Retriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .reasonir_retriever import ReasonIRRetriever
+except ImportError:
+    ReasonIRRetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .reasonembed_retriever import ReasonEmbedRetriever
+except ImportError:
+    ReasonEmbedRetriever = None  # type: ignore[assignment,misc]
+
+try:
+    from .bge_reasoner_retriever import BgeReasonerRetriever
+except ImportError:
+    BgeReasonerRetriever = None  # type: ignore[assignment,misc]
 
 
 __all__ = [
     "Retriever",
-    "BaseRetriever", 
+    "BaseRetriever",
+    "BM25SRetriever",
     "BM25Retriever",
     "DenseRetriever",
-    "ANCERetriever",    # NEW EXPORT
+    "ANCERetriever",
     "BGERetriever",
     "ColBERTRetriever",
     "ContrieverRetriever",

rankify/retrievers/bm25_retriever.py

@@ -1,14 +1,19 @@
 # bm25_retriever.py
 import json
 from typing import List
-from pyserini.search.lucene import LuceneSearcher
-from pyserini.eval.evaluate_dpr_retrieval import has_answers, SimpleTokenizer
 from tqdm import tqdm
 import os
 from .base_retriever import BaseRetriever
 from .index_manager import IndexManager
 from rankify.dataset.dataset import Document, Context
 
+try:
+    from pyserini.search.lucene import LuceneSearcher
+    from pyserini.eval.evaluate_dpr_retrieval import has_answers, SimpleTokenizer
+    _PYSERINI_AVAILABLE = True
+except ImportError:
+    _PYSERINI_AVAILABLE = False
+
 class BM25Retriever(BaseRetriever):
     """
     BM25 retriever implementation using Pyserini's LuceneSearcher.
@@ -17,6 +22,13 @@ class BM25Retriever(BaseRetriever):
     """
 
     def __init__(self, index_type: str = "wiki", index_folder: str = None, **kwargs):
+        if not _PYSERINI_AVAILABLE:
+            raise ImportError(
+                "pyserini is required for BM25Retriever. "
+                "Install it with: pip install pyserini  "
+                "Or use BM25SRetriever for a pure-Python alternative: "
+                "Retriever(method='bm25s', ...)"
+            )
         super().__init__(**kwargs)
         self.index_type = index_type
         self.index_folder = index_folder
@@ -48,7 +60,7 @@ def _load_reverse_mapping(self):
                 fwd = json.load(f)                   # { "orig_id": 123 }
             m = {str(v): k for k, v in fwd.items()}  # ensure string keys
         return m
-    def _initialize_searcher(self) -> LuceneSearcher:
+    def _initialize_searcher(self):
         """Initialize Lucene searcher."""
         if self.index_path.startswith("wikipedia-") or "prebuilt" in self.index_path:
             return LuceneSearcher.from_prebuilt_index(self.index_path)