AF
Ash Framework•4mo ago
marot

How can I return vector_cosine_distance from search results?

I'm currently testing AshAi's embeddings support, and implemented a search based on the docs. I struggle to return the result of vector_cosine_distance(full_text_vector, ^search_vector) for debugging purposes. I tried to use a calculation, but I won't have access to search_vector. In Ecto I'd use a virtual field and use select_merge. What is the correct approach in Ash here? 
read :search do
  argument :query, :string, allow_nil?: false

  prepare before_action(fn query, context ->
    case YourEmbeddingModel.generate([query.arguments.query], []) do
      {:ok, [search_vector]} ->
        Ash.Query.filter(
          query,
          vector_cosine_distance(full_text_vector, ^search_vector) < 0.5
        )
        |> Ash.Query.sort(
          {calc(vector_cosine_distance(full_text_vector, ^search_vector),
             type: :float
           ), :asc}
        )
        |> Ash.Query.limit(10)

      {:error, error} ->
        {:error, error}
    end
  end)
end
read :search do
  argument :query, :string, allow_nil?: false

  prepare before_action(fn query, context ->
    case YourEmbeddingModel.generate([query.arguments.query], []) do
      {:ok, [search_vector]} ->
        Ash.Query.filter(
          query,
          vector_cosine_distance(full_text_vector, ^search_vector) < 0.5
        )
        |> Ash.Query.sort(
          {calc(vector_cosine_distance(full_text_vector, ^search_vector),
             type: :float
           ), :asc}
        )
        |> Ash.Query.limit(10)

      {:error, error} ->
        {:error, error}
    end
  end)
end
Solution:
You can use calculations with arguments, and load that calculation in this preparation
Jump to solution
3 Replies
Solution
ZachDaniel
ZachDaniel•4mo ago
You can use calculations with arguments, and load that calculation in this preparation
ZachDaniel
ZachDaniel•4mo ago
load(vector_cosine_distance: %{search: search})
marot
marotOP•4mo ago
Thanks! For reference: 
      prepare before_action(fn query, _context ->
                search_query = query.arguments.query

                case Sprout.Stories.EmbeddingModel.generate([search_query], []) do
                  {:ok, [search_vector]} ->
                    query
                    |> Ash.Query.filter(
                      vector_cosine_distance(full_text_vector, ^search_vector) < 0.99
                    )
                    |> Ash.Query.sort(vector_distance: {%{search_vector: search_vector}, :asc})
                    |> Ash.Query.limit(20)
                    |> Ash.Query.load(vector_distance: %{search_vector: search_vector})

                  {:error, error} ->
                    {:error, error}
                end
              end)
      prepare before_action(fn query, _context ->
                search_query = query.arguments.query

                case Sprout.Stories.EmbeddingModel.generate([search_query], []) do
                  {:ok, [search_vector]} ->
                    query
                    |> Ash.Query.filter(
                      vector_cosine_distance(full_text_vector, ^search_vector) < 0.99
                    )
                    |> Ash.Query.sort(vector_distance: {%{search_vector: search_vector}, :asc})
                    |> Ash.Query.limit(20)
                    |> Ash.Query.load(vector_distance: %{search_vector: search_vector})

                  {:error, error} ->
                    {:error, error}
                end
              end)
   calculate :vector_distance,
              :float,
              expr(vector_cosine_distance(full_text_vector, ^arg(:search_vector))) do
      description "Calculates the cosine distance between the story's vector and a search vector"

      argument :search_vector, {:array, :float} do
        allow_nil? false
      end

      public? true
    end
   calculate :vector_distance,
              :float,
              expr(vector_cosine_distance(full_text_vector, ^arg(:search_vector))) do
      description "Calculates the cosine distance between the story's vector and a search vector"

      argument :search_vector, {:array, :float} do
        allow_nil? false
      end

      public? true
    end

Did you find this page helpful?