chore: refactor float64 embeddings (#42)

embedder/embedding.go

@@ -5,7 +5,4 @@ var (
 )
 
 // Embedding is the result of an embedding operation.
-type Embedding struct {
-	Embedding []float32 `json:"embedding"`
-	Index     int       `json:"index"`
-}
+type Embedding []float64

embedder/openai/openai.go

@@ -71,11 +71,16 @@ func (t *openAIEmbedder) Embed(ctx context.Context, docs []document.Document) ([
 	var embeddings []embedder.Embedding
 
 	for _, obj := range resp.Data {
-		embeddings = append(embeddings, embedder.Embedding{
-			Embedding: obj.Embedding,
-			Index:     obj.Index,
-		})
+		embeddings = append(embeddings, float32ToFloat64(obj.Embedding))
 	}
 
 	return embeddings, nil
 }
+
+func float32ToFloat64(slice []float32) []float64 {
+	newSlice := make([]float64, len(slice))
+	for i, v := range slice {
+		newSlice[i] = float64(v)
+	}
+	return newSlice
+}

go.mod

@@ -6,8 +6,8 @@ require github.com/mitchellh/mapstructure v1.5.0
 
 require (
 	github.com/google/uuid v1.3.0
-	github.com/henomis/pinecone-go v1.0.3
-	github.com/sashabaranov/go-openai v1.9.0
+	github.com/henomis/pinecone-go v1.1.0
+	github.com/sashabaranov/go-openai v1.9.1
 )
 
 require github.com/henomis/restclientgo v1.0.3 // indirect

go.sum

@@ -1,10 +1,10 @@
 github.com/google/uuid v1.3.0 h1:t6JiXgmwXMjEs8VusXIJk2BXHsn+wx8BZdTaoZ5fu7I=
 github.com/google/uuid v1.3.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
-github.com/henomis/pinecone-go v1.0.3 h1:mhWH2IlfdjZW3rIxz23JHZ3HRUiQaoL5DqVcGq3HnOQ=
-github.com/henomis/pinecone-go v1.0.3/go.mod h1:FsMMRjLyiJ9zHqGOlmGvjolqOp2kkbMsRm8oc85vykU=
+github.com/henomis/pinecone-go v1.1.0 h1:VuiJxTY0wQeUh5e4eD+w30/wKl6KeKYD6292UlsWaOs=
+github.com/henomis/pinecone-go v1.1.0/go.mod h1:FsMMRjLyiJ9zHqGOlmGvjolqOp2kkbMsRm8oc85vykU=
 github.com/henomis/restclientgo v1.0.3 h1:y5+ydfvWJ0/7crObdnCHSn7ya/h1whD+PV4Ir2dZ9Ig=
 github.com/henomis/restclientgo v1.0.3/go.mod h1:xIeTCu2ZstvRn0fCukNpzXLN3m/kRTU0i0RwAbv7Zug=
 github.com/mitchellh/mapstructure v1.5.0 h1:jeMsZIYE/09sWLaz43PL7Gy6RuMjD2eJVyuac5Z2hdY=
 github.com/mitchellh/mapstructure v1.5.0/go.mod h1:bFUtVrKA4DC2yAKiSyO/QUcy7e+RRV2QTWOzhPopBRo=
-github.com/sashabaranov/go-openai v1.9.0 h1:NoiO++IISxxJ1pRc0n7uZvMGMake0G+FJ1XPwXtprsA=
-github.com/sashabaranov/go-openai v1.9.0/go.mod h1:lj5b/K+zjTSFxVLijLSTDZuP7adOgerWeFyZLUhAKRg=
+github.com/sashabaranov/go-openai v1.9.1 h1:3N52HkJKo9Zlo/oe1AVv5ZkCOny0ra58/ACvAxkN3MM=
+github.com/sashabaranov/go-openai v1.9.1/go.mod h1:lj5b/K+zjTSFxVLijLSTDZuP7adOgerWeFyZLUhAKRg=

index/index.go

@@ -20,7 +20,7 @@ const (
 type SearchResponse struct {
 	ID       string
 	Document document.Document
-	Score    float32
+	Score    float64
 }
 
 type Embedder interface {

index/pinecone.go

@@ -132,7 +132,7 @@ func (p *pinecone) similaritySearch(ctx context.Context, topK *int, query string
 			IndexName:       p.indexName,
 			ProjectID:       p.projectID,
 			TopK:            int32(pineconeTopK),
-			Vector:          embeddings[0].Embedding,
+			Vector:          embeddings[0],
 			IncludeMetadata: &includeMetadata,
 			Namespace:       &p.namespace,
 		},
@@ -228,7 +228,7 @@ func buildVectorsFromEmbeddingsAndDocuments(
 
 		vectors = append(vectors, pineconerequest.Vector{
 			ID:       vectorID.String(),
-			Values:   embedding.Embedding,
+			Values:   embedding,
 			Metadata: metadata,
 		})
 
@@ -258,7 +258,7 @@ func buildSearchReponsesFromMatches(matches []pineconeresponse.QueryMatch, inclu
 			id = *match.ID
 		}
 
-		score := float32(0)
+		score := float64(0)
 		if match.Score != nil {
 			score = *match.Score
 		}

index/simpleVectorIndex.go

@@ -142,27 +142,27 @@ func (s *simpleVectorIndex) SimilaritySearch(ctx context.Context, query string,
 	return filterSearchResponses(searchResponses, topK), nil
 }
 
-func (s *simpleVectorIndex) cosineSimilarity(a embedder.Embedding, b embedder.Embedding) float32 {
-	dotProduct := float32(0.0)
-	normA := float32(0.0)
-	normB := float32(0.0)
-
-	for i := 0; i < len(a.Embedding); i++ {
-		dotProduct += a.Embedding[i] * b.Embedding[i]
-		normA += a.Embedding[i] * a.Embedding[i]
-		normB += b.Embedding[i] * b.Embedding[i]
+func (s *simpleVectorIndex) cosineSimilarity(a embedder.Embedding, b embedder.Embedding) float64 {
+	dotProduct := float64(0.0)
+	normA := float64(0.0)
+	normB := float64(0.0)
+
+	for i := 0; i < len(a); i++ {
+		dotProduct += a[i] * b[i]
+		normA += a[i] * a[i]
+		normB += b[i] * b[i]
 	}
 
 	if normA == 0 || normB == 0 {
-		return float32(0.0)
+		return float64(0.0)
 	}
 
-	return dotProduct / (float32(math.Sqrt(float64(normA))) * float32(math.Sqrt(float64(normB))))
+	return dotProduct / (math.Sqrt(normA) * math.Sqrt(normB))
 }
 
-func (s *simpleVectorIndex) cosineSimilarityBatch(a embedder.Embedding) []float32 {
+func (s *simpleVectorIndex) cosineSimilarityBatch(a embedder.Embedding) []float64 {
 
-	scores := make([]float32, len(s.data))
+	scores := make([]float64, len(s.data))
 
 	for i := range s.data {
 		scores[i] = s.cosineSimilarity(a, s.data[i].Embedding)