fernando.melo/itti_adb
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add selectai

Browse Source
This commit is contained in:
Fernando Melo
2026-05-28 14:38:15 -03:00
parent 440b1cfadb
commit 052eb23d54
10 changed files with 581 additions and 112 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
lab/vector/04_indexing.sql
View File

@@ -1,6 +1,6 @@
set long 999999
set pages 1000
set lines 200
set pagesize 1000
set linesize 200
/* --------------------------------
@@ -9,8 +9,6 @@ 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
@@ -22,6 +20,7 @@ PARAMETERS (
EFCONSTRUCTION 200
);
DROP INDEX IF EXISTS markdown_chunks_hnsw_idx;
/* --------------------------------
@@ -30,8 +29,6 @@ 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)
@@ -44,7 +41,7 @@ PARAMETERS (
MIN_VECTORS_PER_PARTITION 1
);
DROP INDEX IF EXISTS markdown_chunks_ivf_idx;
/* --------------------------------
TEXT INDEX + VECTOR SEARCH
@@ -63,24 +60,22 @@ select
SCORE(1) as text_score,
VECTOR_DISTANCE(embedding,
VECTOR_EMBEDDING(MULTILINGUAL_E5_BASE
USING 'estaciones de servicio' as DATA
USING 'Sucursales adheridas' 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
where contains(chunk_text, '(Mastercard ACCUM NEAR((Dúo,Clásica),1))', 1) > 0
order by vector_distance
fetch first 3 rows only;
-- Promoción
-- FUZZY(Promción)
-- ABOUT(Promoción)
-- NEAR((tarjetas, físicas, billeteras, electrónicas), 5)
-- Black OR Albirroja
-- BENDITA ACCUM STYLE
-- (Mastercard ACCUM NEAR((Dúo,Clásica),1))
/* --------------------------------
@@ -103,14 +98,15 @@ PARALLEL 2;
SELECT JSON_SERIALIZE(
DBMS_HYBRID_VECTOR.SEARCH(
JSON('{
"hybrid_index_name" : "markdown_chunks_hybrid_idx",
"hybrid_index_name" : "markdown_chunks_hybrid_idx",
"search_fusion" : "UNION",
"search_scorer" : "rsf",
"vector" : {
"search_text" : "beneficios en estaciones de servicio",
"search_mode" : "CHUNK"
"search_text" : "Vigencia",
"search_mode" : "CHUNK"
},
"text" : {
"contains" : "estaciones AND Petrobras"
"contains" : "Shopping NEAR San NEAR Lorenzo"
},
"return" : {
"values" : [
@@ -135,8 +131,8 @@ FROM
'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'),
'vector' VALUE json_object('search_text' VALUE 'COMERCIOS ADHERIDOS'),
'text' VALUE json_object('contains' VALUE 'BENDITA ACCUM STYLE'),
'return' VALUE json_object(
'values' VALUE json_array('rowid', 'score', 'vector_score', 'vector_rank', 'text_score', 'text_rank', 'chunk_text'),
'topN' VALUE 3