向量分析

前提条件

已根据快速入门，完成了导入向量数据。

向量分析

本教程的向量分析以获取欧氏距离（平方）、点积距离或者余弦相似度为例。

获取欧式距离

使用向量分析，并获取欧氏距离（平方值）。

SELECT id, l2_squared_distance(feature, array[0.495181661387,0.108697291209,0.181728549067,0.109680543346,0.19713082404,0.0197809514512,0.534227452778,0.442411970815,0.409909873031,0.0975687394505]::float4[]) AS distance 
  FROM vector_test.car_info 
  ORDER BY feature <-> array[0.495181661387,0.108697291209,0.181728549067,0.109680543346,0.19713082404,0.0197809514512,0.534227452778,0.442411970815,0.409909873031,0.0975687394505]::float4[] 
  LIMIT 10;

返回示例如下。

  id  |      distance
------+--------------------
    2 |                  0
 1331 | 0.0677967891097069
 1543 |  0.079616591334343
 5606 | 0.0892329216003418
 6423 | 0.0894578248262405
 1667 | 0.0903968289494514
 8215 | 0.0936210229992867
 7801 | 0.0952572822570801
 2581 | 0.0965127795934677
 2645 | 0.0987173467874527
(10 rows)

获取点积距离（余弦相似度）

使用向量分析，并获取点积距离（在归一化时，点积距离等于余弦相似度）。

SELECT id, dp_distance(feature, array[0.495181661387,0.108697291209,0.181728549067,0.109680543346,0.19713082404,0.0197809514512,0.534227452778,0.442411970815,0.409909873031,0.0975687394505]::float4[]) AS similarity 
  FROM vector_test.car_info 
  ORDER BY feature <-> array[0.495181661387,0.108697291209,0.181728549067,0.109680543346,0.19713082404,0.0197809514512,0.534227452778,0.442411970815,0.409909873031,0.0975687394505]::float4[] 
  LIMIT 10;

返回示例如下。

  id  |    similarity
------+-------------------
    2 |                 1
 1331 | 0.966101586818695
 1543 | 0.960191607475281
 5606 | 0.955383539199829
 6423 | 0.955271065235138
 1667 | 0.954801559448242
 8215 | 0.953189492225647
 7801 |  0.95237135887146
 2581 | 0.951743602752686
 2645 | 0.950641334056854
(10 rows)

融合检索查询

如需进行结构化与非结构化的融合，可以采用如下SQL进行查询。

SELECT id, dp_distance(feature, array[0.495181661387,0.108697291209,0.181728549067,0.109680543346,0.19713082404,0.0197809514512,0.534227452778,0.442411970815,0.409909873031,0.0975687394505]::float4[]) AS similarity 
  FROM vector_test.car_info 
  WHERE market_time >= '2020-10-30 00:00:00' 
  AND market_time < '2021-01-01 00:00:00' 
  AND color in ('red', 'white', 'blue') 
  AND price < 100 
  ORDER BY feature <-> array[0.495181661387,0.108697291209,0.181728549067,0.109680543346,0.19713082404,0.0197809514512,0.534227452778,0.442411970815,0.409909873031,0.0975687394505]::float4[] 
  LIMIT 10;

返回示例如下。

  id  |    similarity
------+-------------------
 7645 | 0.922723233699799
 8956 | 0.920517802238464
 8219 |  0.91210675239563
 8503 | 0.895939946174622
 5113 | 0.895431876182556
 7680 | 0.893448948860168
 8433 | 0.893425941467285
 3604 |  0.89293098449707
 3945 | 0.891274154186249
 7153 | 0.891128540039062
(10 rows)

相关文档

向量分析