Skip to content

Open Source alternative to Algolia + Pinecone and an Easier-to-Use alternative to ElasticSearch ⚡ 🔍 ✨ Fast, typo tolerant, in-memory fuzzy Search Engine for building delightful search experiences

Disable bucketing if not enough results are found.

Kishore Nallancd234c306222
diff --git a/src/collection.cpp b/src/collection.cpp
index 27a41b45..d9a35b81 100644
--- a/src/collection.cpp
+++ b/src/collection.cpp
@@ -1234,12 +1234,6 @@ Option<nlohmann::json> Collection::search(const std::string & raw_query,
for(size_t i = 0; i < sort_fields_std.size(); i++) {
if(sort_fields_std[i].name == sort_field_const::text_match && sort_fields_std[i].text_match_buckets != 0) {
match_score_index = i;
-
- if(sort_fields_std[i].text_match_buckets > 1) {
- // we will disable prioritize exact match because it's incompatible with bucketing
- prioritize_exact_match = false;
- }
-
break;
}
}
@@ -1337,31 +1331,33 @@ Option<nlohmann::json> Collection::search(const std::string & raw_query,
if(match_score_index >= 0 && sort_fields_std[match_score_index].text_match_buckets > 1) {
size_t num_buckets = sort_fields_std[match_score_index].text_match_buckets;
-
const size_t max_kvs_bucketed = std::min<size_t>(DEFAULT_TOPSTER_SIZE, raw_result_kvs.size());
- std::vector<int64_t> result_scores(max_kvs_bucketed);
- // only first `max_kvs_bucketed` elements are bucketed to prevent pagination issues past 250 records
- size_t block_len = (max_kvs_bucketed < num_buckets) ? max_kvs_bucketed : (max_kvs_bucketed / num_buckets) + 1;
- size_t i = 0;
- while(i < max_kvs_bucketed) {
- int64_t anchor_score = raw_result_kvs[i][0]->scores[raw_result_kvs[i][0]->match_score_index];
- size_t j = 0;
- while(j < block_len && i+j < max_kvs_bucketed) {
- result_scores[i+j] = raw_result_kvs[i+j][0]->scores[raw_result_kvs[i+j][0]->match_score_index];
- raw_result_kvs[i+j][0]->scores[raw_result_kvs[i+j][0]->match_score_index] = anchor_score;
- j++;
- }
+ if(max_kvs_bucketed >= num_buckets) {
+ std::vector<int64_t> result_scores(max_kvs_bucketed);
- i += j;
- }
+ // only first `max_kvs_bucketed` elements are bucketed to prevent pagination issues past 250 records
+ size_t block_len = (max_kvs_bucketed / num_buckets);
+ size_t i = 0;
+ while(i < max_kvs_bucketed) {
+ int64_t anchor_score = raw_result_kvs[i][0]->scores[raw_result_kvs[i][0]->match_score_index];
+ size_t j = 0;
+ while(j < block_len && i+j < max_kvs_bucketed) {
+ result_scores[i+j] = raw_result_kvs[i+j][0]->scores[raw_result_kvs[i+j][0]->match_score_index];
+ raw_result_kvs[i+j][0]->scores[raw_result_kvs[i+j][0]->match_score_index] = anchor_score;
+ j++;
+ }
- // sort again based on bucketed match score
- std::sort(raw_result_kvs.begin(), raw_result_kvs.end(), Topster::is_greater_kv_group);
+ i += j;
+ }
- // restore original scores
- for(i = 0; i < max_kvs_bucketed; i++) {
- raw_result_kvs[i][0]->scores[raw_result_kvs[i][0]->match_score_index] = result_scores[i];
+ // sort again based on bucketed match score
+ std::sort(raw_result_kvs.begin(), raw_result_kvs.end(), Topster::is_greater_kv_group);
+
+ // restore original scores
+ for(i = 0; i < max_kvs_bucketed; i++) {
+ raw_result_kvs[i][0]->scores[raw_result_kvs[i][0]->match_score_index] = result_scores[i];
+ }
}
}
diff --git a/test/collection_sorting_test.cpp b/test/collection_sorting_test.cpp
index 4de53863..80ffbaf5 100644
--- a/test/collection_sorting_test.cpp
+++ b/test/collection_sorting_test.cpp
@@ -1646,14 +1646,15 @@ TEST_F(CollectionSortingTest, TextMatchBucketRanking) {
spp::sparse_hash_set<std::string>(), 10, "", 30, 4, "title", 20, {}, {}, {}, 0,
"<mark>", "</mark>", {3, 1}, 1000, true).get();
+ // when there are more buckets than results, no bucketing will happen
ASSERT_EQ(2, results["hits"].size());
- ASSERT_EQ("1", results["hits"][0]["document"]["id"].get<std::string>());
- ASSERT_EQ("0", results["hits"][1]["document"]["id"].get<std::string>());
+ ASSERT_EQ("0", results["hits"][0]["document"]["id"].get<std::string>());
+ ASSERT_EQ("1", results["hits"][1]["document"]["id"].get<std::string>());
// bucketing by 1 produces original text match
sort_fields = {
- sort_by("_text_match(buckets: 1)", "DESC"),
- sort_by("points", "DESC"),
+ sort_by("_text_match(buckets: 1)", "DESC"),
+ sort_by("points", "DESC"),
};
results = coll1->search("mark", {"title", "description"},
@@ -1768,8 +1769,8 @@ TEST_F(CollectionSortingTest, TextMatchMoreDocsThanBuckets) {
ASSERT_EQ(4, results["hits"].size());
ASSERT_EQ("3", results["hits"][0]["document"]["id"].get<std::string>());
- ASSERT_EQ("2", results["hits"][1]["document"]["id"].get<std::string>());
- ASSERT_EQ("0", results["hits"][2]["document"]["id"].get<std::string>());
+ ASSERT_EQ("0", results["hits"][1]["document"]["id"].get<std::string>());
+ ASSERT_EQ("2", results["hits"][2]["document"]["id"].get<std::string>());
ASSERT_EQ("1", results["hits"][3]["document"]["id"].get<std::string>());
collectionManager.drop_collection("coll1");