MilvusPlus: Enhanced Operations for Vector Databases

MilvusPlus is a powerful Java library that streamlines interactions with Milvus vector databases, offering an intuitive API for developers familiar with MyBatis-Plus style annotations and method invocations.

Table of Contents

1. Features
2. Getting Started
3. Application Scenarios
4. Annotations
5. Contributing
6. License

Features

• Annotation-based Configuration: Similar to MyBatis-Plus, use annotations to configure your entity models.
• Intuitive API: A straightforward API that makes vector database operations feel natural.
• Extensible: Designed with extensibility in mind, allowing for easy addition of new features.
• Type-Safe: Leverage Java's type safety to minimize errors.

Getting Started

Add MilvusPlus to your project:

Maven:

<dependency>
    <groupId>io.github.javpower</groupId>
    <artifactId>milvus-plus-boot-starter</artifactId>
    <version>2.4.0</version>
</dependency>

Application Scenarios

Vector databases are particularly useful in:

• Similarity Search: Quickly find items most similar to a given vector.
• Recommendation Systems: Recommend content based on user behavior and preferences.
• Image Retrieval: Identify images most similar to a query image in a large database.
• Natural Language Processing: Perform semantic searches by converting text to vectors.
• Bioinformatics: Analyze and compare biological sequences like proteins and genomes.

Annotations

MilvusPlus introduces several annotations to map your Java entities to Milvus collections:

• @MilvusCollection: Denotes a Java class as a Milvus collection.
• @MilvusField: Maps a Java field to a Milvus field with options for data type, dimension, and more.
• @MilvusIndex: Defines an index on a Milvus field.

Example usage:

@Data
@MilvusCollection(name = "face_collection") // 指定Milvus集合的名称
public class Face {
    @MilvusField(
            name = "person_id", // 字段名称
            dataType = DataType.Int64, // 数据类型为64位整数
            isPrimaryKey = true, // 标记为主键
            autoID = true // 假设这个ID是自动生成的
    )
    private Long personId; // 人员的唯一标识符

    @MilvusField(
            name = "face_vector", // 字段名称
            dataType = DataType.FloatVector, // 数据类型为浮点型向量
            dimension = 128, // 向量维度，假设人脸特征向量的维度是128
            isPartitionKey = false // 假设这个字段不是分区键
    )
    @MilvusIndex(
            indexType = IndexParam.IndexType.IVF_FLAT, // 使用IVF_FLAT索引类型
            metricType = IndexParam.MetricType.L2, // 使用L2距离度量类型
            indexName = "face_index", // 索引名称
            extraParams = { // 指定额外的索引参数
                    @ExtraParam(key = "nlist", value = "100") // 例如，IVF的nlist参数
            }
    )
    private List<Float> faceVector; // 存储人脸特征的向量
}

@Component
public class FaceMilvusMapper extends MilvusMapper<Face> {
    
}

@Component
public class ApplicationRunnerTest implements ApplicationRunner {
    @Autowired
    private FaceMilvusMapper mapper;
    @Override
    public void run(ApplicationArguments args) throws Exception {
        List<Float> vector = Lists.newArrayList(0.1f,0.2f,0.3f);

        //查询
        MilvusResp<MilvusResultVo<Face>> query =  mapper.searchWrapper()
                .eq(Face::getPersonId, 1l)
                .vector(vector)
                .limit(100l)
                .query();
        MilvusResp<List<Face>> query2 = mapper.getById(1l);


        //删除
        MilvusResp<DeleteResp> remove= mapper.deleteWrapper()
                .eq(Face::getPersonId, 1l)
                .id(111)
                .remove();
        MilvusResp<DeleteResp> remove2 = mapper.removeById(1l);

        //更新
        Face face=new Face();
        face.setFaceVector(vector);
        MilvusResp<UpsertResp> update = mapper.updateWrapper()
                .eq(Face::getPersonId, 1l)
                .update(face);
        face.setPersonId(1l);
        MilvusResp<UpsertResp> update2 = mapper.updateById(face);
    }
}

Contributing

Contributions are welcome!

• Report issues or suggest features by opening an issue.
• Submit changes by creating a pull request.

License

MilvusPlus is open source and available under the License.

Contact

For questions or support, reach out to javpower@163.com.