FunctionChain

An ordered rerank/refine plan applied to search results. A FunctionChain composes map, sort, and limit operations that post-process the result columns of a search() call.

Java

io.milvus.v2.service.vector.request.FunctionChain

Constructor

This builder composes a function chain from map, sort, and limit operations.

Java

FunctionChain.builder()
    .stage(FunctionChainStage stage)
    .name(String name)
    .map(String output, FunctionChainExpr expr)
    .sort(String by, boolean desc, String tieBreakCol)
    .limit(int limit)
    .limit(int limit, int offset)
    .build()

BUILDER METHODS:

• stage(FunctionChainStage stage)

  The execution stage where this function chain runs. Throws MilvusClientException if the stage is null. See FunctionChainStage.

• name(String name)

  The name of this function chain. An empty string is used when null is passed.

• map(String output, FunctionChainExpr expr)

  Adds a map operation that computes a new output column from an expression. See FunctionChainExpr.

• sort(String by, boolean desc, String tieBreakCol)

  Adds a sort operation that sorts the results by the given column. Pass null for tieBreakCol to omit the tie-break column.

• limit(int limit)

  Adds a limit operation that keeps only the given number of results, starting from the beginning.

• limit(int limit, int offset)

  Adds a limit operation that keeps only the given number of results, skipping the given number of leading results.

RETURN TYPE:

FunctionChain

RETURNS:

A FunctionChain instance.

METHODS:

• FunctionChainStage getStage()

  Returns the execution stage where this function chain runs.

• String getName()

  Returns the name of this function chain.

• List<FunctionChainOp> getOps()

  Returns the ordered list of operations of this function chain.

Example

Java

FunctionChain chain = FunctionChain.builder()
        .stage(FunctionChainStage.L2_RERANK)
        .name("fresh_popular_rerank")
        .map("$score", FunctionChainExpr.builder()
                .name("num_combine")
                .arg(FunctionChainArg.col("$score"))
                .arg(FunctionChainArg.col("freshness"))
                .param("mode", "weighted")
                .param("weights", Arrays.asList(0.7, 0.2, 0.1))
                .build())
        .sort("$score", true, "$id")
        .limit(10)
        .build();

SearchResp response = client.search(SearchReq.builder()
        .collectionName("my_collection")
        .data(Collections.singletonList(queryVector))
        .annsField("vector")
        .functionChains(Collections.singletonList(chain))
        .build());