How to MapReduce ?

[jc3ds]

To follow up on issue #508, it would be great to have the capability to do some kind of MapReduce on arrays of strings.

Scenarios like WebSearch returning more than one result would benefit from

• mapping each result to a function (e.g. for summarizing, checking, re-ranking, ...)
• reducing the outputs to yield a single prompt

According to the documentation: "Branching features such as "if", "for", and code blocks are not part of SK's template language. This reflects SK's design principle of using natural language as much as possible, with a clear separation from conventional programming code."

So, what would be the recommended way to implement such a scenario?

[evchaki]

@jc3ds for a small amount of data you can use function chaining to map and filter data. For larger scenarios we recommend using a native function for processing and then result the filtered results.

[a-bormark]

What do you mean by "native function"?

[Rehu199]

To follow up on issue , it would be great to have the capability to do some kind of MapReduce on arrays of strings.

Scenarios like WebSearch returning more than one result would benefit from

• mapping each result to a function (e.g. for summarizing, checking, re-ranking, ...)
• reducing the outputs to yield a single prompt

According to the documentation: "Branching features such as "if", "for", and code blocks are not part of SK's template language. This reflects SK's design principle of using natural language as much as possible, with a clear separation from conventional programming code."

So, what would be the recommended way to implement such a scenario?

[anthonypuppo]

The extension method AggregatePartitionedResultsAsync is effectively a map reduce. It will apply the same function to a partitioned data set (in this case, an array of strings). You can then further "reduce" if need be or just display the aggregated result.

semantic-kernel/dotnet/src/SemanticKernel.Core/Functions/SKFunctionTextExtensions.cs

Line 29 in ef4c5a5

public static async Task<SKContext> AggregatePartitionedResultsAsync(