[ElasticStack-7] Dense Vector in Elasticsearch

Elasticsearch에서는 vector 값을 직접 입력할 수 있다. dense_vector로 불리는 이 필드는 float 타입의 데이터를 저장 가능하다. (A dense_vector field stores dense vectors of float values.)

필드는 차원 값을 가지는데 2048을 초과할 수 없다. (The maximum number of dimensions that can be in a vector should not exceed 2048.  A dense_vector field is a single-valued field.)

 

Example in Dev tools

## Create index and mapping
PUT dense_vector_test_index
{
	"mappings": {
    	"properties": {
        	"my_dense_vector": {
            	"type" : "dense_vector",
                "dims" : 100
            },
            "my_test_text" : {
            	"type" : "keyword"
            }
        }
    }
}

## Input example vector value
PUT dense_vector_test_index/_doc/3
{	## All data in Elasticsearch, type is doc. 3 is temporary ID in doc.
	"my_test_text" : "corporation",
    "my_dense_vector" : [ 0.30044875, -08096953, 0.03830861, 0.02427242, 0.05239175,
    			0.11283469, 0.01669317, 0.07109312, -0.10938835, -0.17000835,
                0.16206388, 0.18072662, -0.11470463, 0.01919608, -0.04658141,
                0.05263321, 0.09595548, 0.02676269, 0.01009388, -0.6520367,
                ...,
                ...,
                ...,
                ] # You can insert values 100 counts.    
}

### Result example
{
...
{
	"_index" : "dense_vector_test_index",
    "_type" : "_doc",
    "_id" : "1",
    "_score" : 1.0,
    "_source" : {
    	"my_test_text" : "plot",
        "my_dense_vector" : " { ... }
    }
}
...
}

참고 Link

Dense vector datatype | Elasticsearch Reference [7.6] | Elastic