Normalize the vectors during indexation and search

milli/src/lib.rs

@@ -285,6 +285,18 @@ pub fn normalize_facet(original: &str) -> String {
     CompatibilityDecompositionNormalizer.normalize_str(original.trim()).to_lowercase()
 }
 
+/// Normalize a vector by dividing the dimensions by the lenght of it.
+pub fn normalize_vector(mut vector: Vec<f32>) -> Vec<f32> {
+    let squared: f32 = vector.iter().map(|x| x * x).sum();
+    let length = squared.sqrt();
+    if length <= f32::EPSILON {
+        vector
+    } else {
+        vector.iter_mut().for_each(|x| *x = *x / length);
+        vector
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use serde_json::json;

milli/src/search/new/mod.rs

@@ -49,7 +49,8 @@ use self::interner::Interned;
 use crate::score_details::{ScoreDetails, ScoringStrategy};
 use crate::search::new::distinct::apply_distinct_rule;
 use crate::{
-    AscDesc, DocumentId, Filter, Index, Member, Result, TermsMatchingStrategy, UserError, BEU32,
+    normalize_vector, AscDesc, DocumentId, Filter, Index, Member, Result, TermsMatchingStrategy,
+    UserError, BEU32,
 };
 
 /// A structure used throughout the execution of a search query.
@@ -454,7 +455,8 @@ pub fn execute_search(
             let hnsw = ctx.index.vector_hnsw(ctx.txn)?.unwrap_or_default();
             let ef = hnsw.len().min(100);
             let mut dest = vec![Neighbor { index: 0, distance: 0 }; ef];
-            let neighbors = hnsw.nearest(vector, ef, &mut searcher, &mut dest[..]);
+            let vector = normalize_vector(vector.clone());
+            let neighbors = hnsw.nearest(&vector, ef, &mut searcher, &mut dest[..]);
 
             let mut docids = Vec::new();
             for Neighbor { index, distance: _ } in neighbors.iter() {

milli/src/update/index_documents/extract/mod.rs

@@ -283,7 +283,7 @@ fn send_and_extract_flattened_documents_data(
     faceted_fields: &HashSet<FieldId>,
     primary_key_id: FieldId,
     geo_fields_ids: Option<(FieldId, FieldId)>,
-    vector_field_id: Option<FieldId>,
+    vectors_field_id: Option<FieldId>,
     stop_words: &Option<fst::Set<&[u8]>>,
     max_positions_per_attributes: Option<u32>,
 ) -> Result<(
@@ -312,11 +312,11 @@ fn send_and_extract_flattened_documents_data(
         });
     }
 
-    if let Some(vector_field_id) = vector_field_id {
+    if let Some(vectors_field_id) = vectors_field_id {
         let documents_chunk_cloned = flattened_documents_chunk.clone();
         let lmdb_writer_sx_cloned = lmdb_writer_sx.clone();
         rayon::spawn(move || {
-            let result = extract_vector_points(documents_chunk_cloned, indexer, vector_field_id);
+            let result = extract_vector_points(documents_chunk_cloned, indexer, vectors_field_id);
             let _ = match result {
                 Ok(vector_points) => {
                     lmdb_writer_sx_cloned.send(Ok(TypedChunk::VectorPoints(vector_points)))

milli/src/update/index_documents/typed_chunk.rs

@@ -19,6 +19,7 @@ use super::helpers::{
 use super::{ClonableMmap, MergeFn};
 use crate::error::UserError;
 use crate::facet::FacetType;
+use crate::normalize_vector;
 use crate::update::facet::FacetsUpdate;
 use crate::update::index_documents::helpers::{as_cloneable_grenad, try_split_array_at};
 use crate::{lat_lng_to_xyz, CboRoaringBitmapCodec, DocumentId, GeoPoint, Index, Result, BEU32};
@@ -253,6 +254,7 @@ pub(crate) fn write_typed_chunk_into_index(
                     return Err(UserError::InvalidVectorDimensions { expected, found })?;
                 }
 
+                let vector = normalize_vector(vector);
                 let vector_id = hnsw.insert(vector, &mut searcher) as u32;
                 index.vector_id_docid.put(wtxn, &BEU32::new(vector_id), &BEU32::new(docid))?;
             }