python读写es: mapping类型的index

目录

常规index

mapping类型的index

插入1条向量

计算相似度

分值翻译

封装成类

---

常规index

from elasticsearch import Elasticsearch

es = Elasticsearch('http://xx:9200')
print(es.indices.create(index='test01'))

mapping类型的index

import requests
from elasticsearch import Elasticsearch

es = Elasticsearch('http://xx:9200')
mapping = {'properties':{
                'vector':{
                    'type': 'dense_vector',
                    'dims': 3,
                    'index': True,
                    'similarity': 'dot_product'
                    }
                }
}
print(es.indices.create(index='test03', mappings=mapping))

插入1条向量

import requests
from elasticsearch import Elasticsearch

es = Elasticsearch('http://xxx:9200')

res = es.index(index='test03',document={'vector': [0.1,0.1,0.98995]})
print(res)

{'_index': 'test03', '_id': 'h_v5TYkBNiG-_5JLvCS6', '_version': 1, 'result': 'created', '_shards': {'total': 2, 'successful': 1, 'failed': 0}, '_seq_no': 0, '_primary_term': 1}

计算相似度

import requests
from elasticsearch import Elasticsearch

es = Elasticsearch('http://xx:9200')

knn = {'field': 'vector',
       'query_vector': [0.1,0.98995,0.1],
       'k': 1,
       'num_candidates': 100}

res = es.search(index='test03',knn=knn)
print(res)

{'took': 1, 'timed_out': False, '_shards': {'total': 1, 'successful': 1, 'skipped': 0, 'failed': 0}, 'hits': {'total': {'value': 1, 'relation': 'eq'}, 'max_score': 0.60399497, 'hits': [{'_index': 'test03', '_id': 'h_v5TYkBNiG-_5JLvCS6', '_score': 0.60399497, '_source': {'vector': [0.1, 0.1, 0.98995]}}]}}

分值翻译

参考：https://www.elastic.co/guide/en/elasticsearch/reference/8.6/dense-vector.html

点积值 = (es返回的score)x2 -1

• [0.1,0.98995,0.1]

• [0.1,0.1,0.98995]

a = [0.1,0.1,0.98995]
b = [0.1,0.98995,0.1]

IP = 0
for i,j in zip(a,b):
    IP += i*j
print(IP)

es_score = 0.60399497
print(2*es_score-1)

0.20799
0.20798994000000004

封装成类

class EsVector:
    def __init__(self):
        self.es = Elasticsearch('http://xx:9200')
        self.indexName = 'test03'
        mapping = {'properties':{
                      'vector':{
                         'type': 'dense_vector',
                         'dims': 512,
                         'index': True,
                         'similarity': 'dot_product'}}}
        if self.es.indices.exists(index=self.indexName):
            self.es.indices.delete(index=self.indexName)
            self.es.indices.create(index=self.indexName, mappings=mapping)
            print('delete index={}. create index={}'.format(self.indexName,self.indexName))
        else:
            self.es.indices.create(index=self.indexName, mappings=mapping)
            print('create index={}'.format(self.indexName))

    def vectorNum(self):
        n = self.es.count(index=self.indexName)['count']
        return n

    def close(self):
        self.es.close()

    def insert(self,vector):
        res = self.es.index(index=self.indexName, document={'vector': vector[0]})
        self.es.indices.refresh(index=self.indexName)
        return res['_id']

    def search(self,vector):
        knn = {'field': 'vector', 'query_vector': None, 'k': 1, 'num_candidates': 100}
        knn['query_vector'] = vector[0]
        res = self.es.search(index=self.indexName, knn=knn)
        id = res['hits']['hits'][0]['_id']
        ip = res['hits']['hits'][0]['_score']
        true_ip = 2 * ip -1
        return [id,true_ip]