flexvec 0.2.0__tar.gz → 0.3.0__tar.gz

{flexvec-0.2.0 → flexvec-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: flexvec
-Version: 0.2.0
+Version: 0.3.0
 Summary: numpy-backed semantic search for any SQLite database
 Author-email: Damian Delmas <damian@getflex.dev>
 License: MIT

{flexvec-0.2.0 → flexvec-0.3.0}/flexvec/score.py

@@ -32,13 +32,10 @@ def parse_modifiers(modifier_str: str) -> dict:
         from:TEXT to:TEXT    trajectory — direction through embedding space
         pool:N               candidate count (default 500)
         communities          per-query Louvain, adds _community
-        pagerank             PageRank on candidate subgraph, adds _local_centrality
-        peaks                HDBSCAN cluster extremes, adds _cluster_id, _is_attractor
 
     Deprecated aliases (accepted, will be removed):
         like: → centroid:, unlike: → suppress:, limit: → pool:, recent: → decay:
         local_communities → communities, detect_communities → communities
-        local_pagerank → pagerank
 
     Dead tokens (silently ignored): kind:TYPE, community:N
     Unknown tokens silently ignored (forward-compatible).
@@ -54,10 +51,6 @@ def parse_modifiers(modifier_str: str) -> dict:
         'trajectory_from': None,
         'trajectory_to': None,
         'local_communities': False,
-        'local_pagerank': False,
-        
-        
-        'peaks': False,
     }
 
     if not modifier_str:
@@ -73,13 +66,10 @@ def parse_modifiers(modifier_str: str) -> dict:
     modifier_str = re.sub(r'\brecent\b', 'decay', modifier_str)
     modifier_str = modifier_str.replace('local_communities', 'communities')
     modifier_str = modifier_str.replace('detect_communities', 'communities')
-    modifier_str = modifier_str.replace('local_pagerank', 'pagerank')
-
     # Known token prefixes for boundary detection (canonical names only)
     _TOKEN_BOUNDARY = (
         r'diverse|decay:|suppress:|centroid:|pool:|'
-        r'communities|pagerank|'
-        r'peaks|from:|similar:'
+        r'communities|from:|similar:'
     )
 
     # Extract similar:TEXT (multi-word, up to next token boundary)
@@ -138,10 +128,6 @@ def parse_modifiers(modifier_str: str) -> dict:
             result['like'] = token.split(':', 1)[1].split(',')
         elif token == 'communities':
             result['local_communities'] = True
-        elif token == 'pagerank':
-            result['local_pagerank'] = True
-        elif token == 'peaks':
-            result['peaks'] = True
         # kind: and community: silently ignored (dead tokens)
 
     return result
@@ -375,53 +361,6 @@ def score_candidates(
                     enrichment[int(node)] = {'_community': ci}
             _merge_enrichment(enrichment)
 
-    # Local PageRank on candidate subgraph
-    if modifiers and modifiers.get('local_pagerank') and len(cand_indices) >= 3:
-        import networkx as nx
-        sims = cand_vecs @ cand_vecs.T
-        rows, cols = np.where(np.triu(sims > 0.3, k=1))
-        G = nx.Graph()
-        G.add_nodes_from(range(len(cand_indices)))
-        G.add_weighted_edges_from(
-            (int(r), int(c), float(sims[r, c])) for r, c in zip(rows, cols)
-        )
-        if G.number_of_edges() > 0:
-            pr = nx.pagerank(G, weight='weight')
-            enrichment = {int(node): {'_local_centrality': float(val)}
-                          for node, val in pr.items()}
-            _merge_enrichment(enrichment)
-
-    # MST — minimum spanning tree spine, leaves, hubs
-
-
-    if modifiers and modifiers.get('peaks') and len(cand_indices) >= 20:
-        try:
-            import hdbscan
-            clusterer = hdbscan.HDBSCAN(
-                min_cluster_size=5, min_samples=3, metric='euclidean')
-            cluster_labels = clusterer.fit_predict(cand_vecs)
-
-            enrichment = {}
-            # Per cluster, find the point farthest from centroid (strange attractor)
-            unique_labels = set(cluster_labels)
-            unique_labels.discard(-1)  # noise
-            attractors = set()
-            for cl in unique_labels:
-                members = np.where(cluster_labels == cl)[0]
-                centroid = cand_vecs[members].mean(axis=0)
-                dists = np.linalg.norm(cand_vecs[members] - centroid, axis=1)
-                attractor_pos = members[np.argmax(dists)]
-                attractors.add(int(attractor_pos))
-
-            for pos in range(len(cand_ids)):
-                enrichment[pos] = {
-                    '_cluster_id': int(cluster_labels[pos]) if cluster_labels[pos] >= 0 else None,
-                    '_is_attractor': 1 if pos in attractors else 0,
-                }
-            _merge_enrichment(enrichment)
-        except ImportError:
-            pass  # hdbscan not installed, silently skip
-
     # === Apply structural enrichments to results ===
     def _attach_enrichments(results_list):
         """Attach _-prefixed structural columns to result dicts."""

{flexvec-0.2.0 → flexvec-0.3.0}/flexvec/vec_ops.py

@@ -288,8 +288,8 @@ def materialize_vec_ops(db, sql: str) -> str:
 
     # Populate temp table (unique name per call for HTTP concurrency)
     # Dynamic column construction: discover all _-prefixed columns from
-    # structural tokens (local_communities, mst, homology, fingerprint, etc.)
-    # and build the schema automatically. Any token can emit any column.
+    # structural tokens (e.g. local_communities) and build the schema
+    # automatically. Any token can emit any column.
     tmp_name = f"_vec_results_{uuid.uuid4().hex[:8]}"
 
     base_cols = [('id', 'TEXT PRIMARY KEY'), ('score', 'REAL')]

{flexvec-0.2.0 → flexvec-0.3.0}/flexvec.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: flexvec
-Version: 0.2.0
+Version: 0.3.0
 Summary: numpy-backed semantic search for any SQLite database
 Author-email: Damian Delmas <damian@getflex.dev>
 License: MIT

{flexvec-0.2.0 → flexvec-0.3.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "flexvec"
-version = "0.2.0"
+version = "0.3.0"
 description = "numpy-backed semantic search for any SQLite database"
 license = { text = "MIT" }
 authors = [{name = "Damian Delmas", email = "damian@getflex.dev"}]