add labs

lab/vector/04_indexing.sql

@@ -0,0 +1,154 @@
+set long 999999
+set pages 1000
+set lines 200
+
+/* --------------------------------
+
+HNSW INDEX 
+
+-------------------------------- */ 
+
+-- crear indice HNSW para busquedas vectoriales aproximadas
+DROP INDEX IF EXISTS markdown_chunks_hnsw_idx;
+
+CREATE VECTOR INDEX markdown_chunks_hnsw_idx
+ON markdown_chunks (embedding)
+ORGANIZATION INMEMORY NEIGHBOR GRAPH
+DISTANCE COSINE
+WITH TARGET ACCURACY 90
+PARAMETERS (
+    TYPE HNSW,
+    NEIGHBORS 32,
+    EFCONSTRUCTION 200
+);
+
+
+/* --------------------------------
+
+IVF INDEX 
+
+-------------------------------- */ 
+
+-- crear indice IVF con DOC_ID incluido para filtros por documento
+DROP INDEX IF EXISTS markdown_chunks_ivf_idx;
+
+CREATE VECTOR INDEX markdown_chunks_ivf_idx
+ON markdown_chunks (embedding)
+INCLUDE (doc_id)
+ORGANIZATION NEIGHBOR PARTITIONS
+DISTANCE COSINE
+WITH TARGET ACCURACY 90
+PARAMETERS (
+    TYPE IVF,
+    NEIGHBOR PARTITIONS 4,
+    MIN_VECTORS_PER_PARTITION 1
+);
+
+
+/* --------------------------------
+
+TEXT INDEX + VECTOR SEARCH
+
+-------------------------------- */ 
+-- hybrid indexes / text indexes para busquedas textuales y combinadas
+DROP INDEX IF EXISTS markdown_chunks_text_idx;
+
+CREATE SEARCH INDEX markdown_chunks_text_idx
+ON markdown_chunks (chunk_text)
+FOR TEXT;
+
+select
+    doc_id,
+    chunk_id,
+    SCORE(1) as text_score,
+    VECTOR_DISTANCE(embedding, 
+                    VECTOR_EMBEDDING(MULTILINGUAL_E5_BASE 
+                                    USING 'estaciones de servicio' as DATA
+                                    )
+                    ) as vector_distance,
+    chunk_text    
+from markdown_chunks
+where contains(chunk_text, 'Beneficio 
+                            AND FUZZY(Benefcio) 
+                            AND ABOUT(estaciones)
+                            AND (Visa accum Mastercard)
+                            AND (Clásica OR Oro)
+                            AND (crédito NOT débito) 
+                            AND NEAR((POS, Infonet), 5)
+                            AND NEAR((App, Premmia, Petrobras), 1)
+                            AND (consumo AND personal)', 1) > 0
+order by vector_distance
+fetch first 3 rows only;
+
+
+
+/* --------------------------------
+
+HYBRID VECTOR INDEX
+
+-------------------------------- */ 
+
+-- crear indice hibrido para combinar busqueda textual y semantica
+DROP INDEX IF EXISTS markdown_chunks_hybrid_idx;
+
+CREATE HYBRID VECTOR INDEX markdown_chunks_hybrid_idx
+ON markdown_chunks (chunk_text)
+PARAMETERS ('MODEL MULTILINGUAL_E5_BASE
+             VECTOR_IDXTYPE HNSW
+             MEMORY 128M')
+PARALLEL 2;
+
+
+-- ejemplo de consulta usando el indice hibrido
+SELECT JSON_SERIALIZE(
+    DBMS_HYBRID_VECTOR.SEARCH(
+        JSON('{
+            "hybrid_index_name" : "markdown_chunks_hybrid_idx",            
+            "search_fusion"     : "UNION",
+            "vector" : {
+                "search_text"  : "beneficios en estaciones de servicio",
+                "search_mode"  : "CHUNK"
+            },
+            "text" : {
+                "contains"     : "estaciones AND Petrobras"
+            },
+            "return" : {
+                "values" : [
+                    "chunk_id",
+                    "chunk_text",
+                    "score",
+                    "vector_score",
+                    "text_score"
+                ],
+                "topN" : 5
+            }
+        }')
+    ) RETURNING CLOB PRETTY
+) AS hybrid_results;
+
+
+SELECT jt.*
+FROM
+   JSON_TABLE(
+        dbms_hybrid_vector.search(
+                json_object(
+                  'hybrid_index_name' VALUE 'markdown_chunks_hybrid_idx',      
+                  'search_fusion'     VALUE 'INTERSECT',
+                  'search_scorer'     VALUE 'rsf',
+                  'vector'            VALUE json_object('search_text' VALUE 'beneficios en estaciones de servicio'), 
+                  'text'              VALUE json_object('contains'    VALUE 'estaciones AND Petrobras'),
+                  'return' VALUE json_object(
+                                  'values' VALUE json_array('rowid', 'score', 'vector_score', 'vector_rank', 'text_score', 'text_rank', 'chunk_text'),
+                                  'topN'   VALUE 3
+                                )
+                  RETURNING JSON
+                )
+            ),
+        '$[*]' COLUMNS idx for ORDINALITY,
+                score        NUMBER PATH '$.score',                
+                vector_score NUMBER PATH '$.vector_score',                
+                vector_rank  NUMBER PATH '$.vector_rank',
+                text_score   NUMBER PATH '$.text_score',
+                text_rank    NUMBER PATH '$.text_rank',
+                chunk_text   VARCHAR2(4000) PATH '$.chunk_text'
+    ) jt;