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

Expose max_candidates parameter to the API.

Kishore Nallan609ac35c6d41
diff --git a/include/collection.h b/include/collection.h
index 6a273710..28d45334 100644
--- a/include/collection.h
+++ b/include/collection.h
@@ -394,7 +394,8 @@ public:
size_t search_stop_millis = 6000*1000,
size_t min_len_1typo = 4,
size_t min_len_2typo = 7,
- bool split_join_tokens = true) const;
+ bool split_join_tokens = true,
+ size_t max_candidates = 4) const;
Option<bool> get_filter_ids(const std::string & simple_filter_query,
std::vector<std::pair<size_t, uint32_t*>>& index_ids);
diff --git a/include/index.h b/include/index.h
index 6b5edc19..32fbf52b 100644
--- a/include/index.h
+++ b/include/index.h
@@ -281,6 +281,7 @@ struct search_args {
size_t search_cutoff_ms;
size_t min_len_1typo;
size_t min_len_2typo;
+ size_t max_candidates;
spp::sparse_hash_set<uint64_t> groups_processed;
std::vector<std::vector<art_leaf*>> searched_queries;
@@ -305,7 +306,8 @@ struct search_args {
const std::vector<const override_t*>& dynamic_overrides,
size_t search_cutoff_ms,
size_t min_len_1typo,
- size_t min_len_2typo):
+ size_t min_len_2typo,
+ size_t max_candidates):
field_query_tokens(field_query_tokens),
search_fields(search_fields), filters(filters), facets(facets),
included_ids(included_ids), excluded_ids(excluded_ids), sort_fields_std(sort_fields_std),
@@ -316,7 +318,7 @@ struct search_args {
prioritize_exact_match(prioritize_exact_match), all_result_ids_len(0),
exhaustive_search(exhaustive_search), concurrency(concurrency),
filter_overrides(dynamic_overrides), search_cutoff_ms(search_cutoff_ms),
- min_len_1typo(min_len_1typo), min_len_2typo(min_len_2typo) {
+ min_len_1typo(min_len_1typo), min_len_2typo(min_len_2typo), max_candidates(max_candidates) {
const size_t topster_size = std::max((size_t)1, max_hits); // needs to be atleast 1 since scoring is mandatory
topster = new Topster(topster_size, group_limit);
@@ -503,7 +505,8 @@ private:
size_t typo_tokens_threshold,
bool exhaustive_search,
size_t min_len_1typo,
- size_t min_len_2typo) const;
+ size_t min_len_2typo,
+ size_t max_candidates) const;
void search_candidates(const uint8_t & field_id,
bool field_is_array,
@@ -688,7 +691,8 @@ public:
size_t concurrency,
size_t search_cutoff_ms,
size_t min_len_1typo,
- size_t min_len_2typo) const;
+ size_t min_len_2typo,
+ size_t max_candidates) const;
Option<uint32_t> remove(const uint32_t seq_id, const nlohmann::json & document, const bool is_update);
diff --git a/src/collection.cpp b/src/collection.cpp
index a326b15e..1b00130a 100644
--- a/src/collection.cpp
+++ b/src/collection.cpp
@@ -678,7 +678,8 @@ Option<nlohmann::json> Collection::search(const std::string & raw_query, const s
const size_t search_stop_millis,
const size_t min_len_1typo,
const size_t min_len_2typo,
- bool split_join_tokens) const {
+ bool split_join_tokens,
+ const size_t max_candidates) const {
std::shared_lock lock(mutex);
@@ -994,7 +995,7 @@ Option<nlohmann::json> Collection::search(const std::string & raw_query, const s
group_by_fields, group_limit, default_sorting_field, prioritize_exact_match,
exhaustive_search, 4, filter_overrides,
search_stop_millis,
- min_len_1typo, min_len_2typo);
+ min_len_1typo, min_len_2typo, max_candidates);
index->run_search(search_params);
diff --git a/src/collection_manager.cpp b/src/collection_manager.cpp
index c3e0eae1..c2322b8c 100644
--- a/src/collection_manager.cpp
+++ b/src/collection_manager.cpp
@@ -565,6 +565,7 @@ Option<bool> CollectionManager::do_search(std::map<std::string, std::string>& re
const char *PINNED_HITS = "pinned_hits";
const char *HIDDEN_HITS = "hidden_hits";
const char *ENABLE_OVERRIDES = "enable_overrides";
+ const char *MAX_CANDIDATES = "max_candidates";
// strings under this length will be fully highlighted, instead of showing a snippet of relevant portion
const char *SNIPPET_THRESHOLD = "snippet_threshold";
@@ -828,6 +829,10 @@ Option<bool> CollectionManager::do_search(std::map<std::string, std::string>& re
req_params[ENABLE_OVERRIDES] = "true";
}
+ if(req_params.count(MAX_CANDIDATES) == 0) {
+ req_params[MAX_CANDIDATES] = exhaustive_search ? "10000" : "4";
+ }
+
bool enable_overrides = (req_params[ENABLE_OVERRIDES] == "true");
CollectionManager & collectionManager = CollectionManager::get_instance();
@@ -892,7 +897,8 @@ Option<bool> CollectionManager::do_search(std::map<std::string, std::string>& re
static_cast<size_t>(std::stol(req_params[SEARCH_CUTOFF_MS])),
static_cast<size_t>(std::stol(req_params[MIN_LEN_1TYPO])),
static_cast<size_t>(std::stol(req_params[MIN_LEN_2TYPO])),
- split_join_tokens
+ split_join_tokens,
+ static_cast<size_t>(std::stol(req_params[MAX_CANDIDATES]))
);
uint64_t timeMillis = std::chrono::duration_cast<std::chrono::milliseconds>(
diff --git a/src/index.cpp b/src/index.cpp
index ee2c5cc6..4f33b5ee 100644
--- a/src/index.cpp
+++ b/src/index.cpp
@@ -1629,7 +1629,8 @@ void Index::run_search(search_args* search_params) {
search_params->concurrency,
search_params->search_cutoff_ms,
search_params->min_len_1typo,
- search_params->min_len_2typo);
+ search_params->min_len_2typo,
+ search_params->max_candidates);
}
void Index::collate_included_ids(const std::vector<std::string>& q_included_tokens,
@@ -1974,7 +1975,7 @@ bool Index::check_for_overrides(const token_ordering& token_order, const string&
search_field(0, window_tokens, search_tokens, nullptr, 0, num_toks_dropped, field_it->second, field_name,
nullptr, 0, {}, {}, 2, searched_queries, topster, groups_processed,
&result_ids, result_ids_len, field_num_results, 0, group_by_fields,
- false, 4, query_hashes, token_order, false, 0, 1, false, 3, 7);
+ false, 4, query_hashes, token_order, false, 0, 1, false, 3, 7, 4);
delete [] result_ids;
@@ -2030,7 +2031,8 @@ void Index::search(std::vector<query_tokens_t>& field_query_tokens,
const size_t concurrency,
const size_t search_cutoff_ms,
size_t min_len_1typo,
- size_t min_len_2typo) const {
+ size_t min_len_2typo,
+ const size_t max_candidates) const {
search_begin = std::chrono::high_resolution_clock::now();
search_stop_ms = search_cutoff_ms;
@@ -2258,7 +2260,7 @@ void Index::search(std::vector<query_tokens_t>& field_query_tokens,
field_num_results, group_limit, group_by_fields, prioritize_exact_match, concurrency,
query_hashes, token_order, field_prefix,
drop_tokens_threshold, typo_tokens_threshold, exhaustive_search,
- min_len_1typo, min_len_2typo);
+ min_len_1typo, min_len_2typo, max_candidates);
} else if(actual_filter_ids_length != 0) {
// indicates exact match query
curate_filtered_ids(filters, curated_ids, exclude_token_ids,
@@ -2306,7 +2308,7 @@ void Index::search(std::vector<query_tokens_t>& field_query_tokens,
field_num_results, group_limit, group_by_fields, prioritize_exact_match, concurrency,
query_hashes, token_order, field_prefix,
drop_tokens_threshold, typo_tokens_threshold, exhaustive_search,
- min_len_1typo, min_len_2typo);
+ min_len_1typo, min_len_2typo, max_candidates);
}
}
@@ -2702,7 +2704,7 @@ void Index::compute_facet_infos(const std::vector<facet>& facets, facet_query_t&
facet_field, facet_field.faceted_name(),
all_result_ids, all_result_ids_len, {}, {}, 2, searched_queries, topster, groups_processed,
&field_result_ids, field_result_ids_len, field_num_results, 0, group_by_fields,
- false, 4, query_hashes, MAX_SCORE, true, 0, 1, false, 3, 1000);
+ false, 4, query_hashes, MAX_SCORE, true, 0, 1, false, 3, 1000, 4);
//LOG(INFO) << "searched_queries.size: " << searched_queries.size();
@@ -2974,7 +2976,8 @@ void Index::search_field(const uint8_t & field_id,
const size_t typo_tokens_threshold,
const bool exhaustive_search,
size_t min_len_1typo,
- size_t min_len_2typo) const {
+ size_t min_len_2typo,
+ const size_t max_candidates) const {
// NOTE: `query_tokens` preserve original tokens, while `search_tokens` could be a result of dropped tokens
@@ -3013,7 +3016,6 @@ void Index::search_field(const uint8_t & field_id,
long long int N = std::accumulate(token_to_costs.begin(), token_to_costs.end(), 1LL, product);
const size_t combination_limit = exhaustive_search ? Index::COMBINATION_MAX_LIMIT : Index::COMBINATION_MIN_LIMIT;
- const size_t num_fuzzy_candidates = exhaustive_search ? 10000 : 4;
while(n < N && n < combination_limit) {
RETURN_CIRCUIT_BREAKER
@@ -3051,7 +3053,7 @@ void Index::search_field(const uint8_t & field_id,
// need less candidates for filtered searches since we already only pick tokens with results
art_fuzzy_search(search_index.at(field_name), (const unsigned char *) token.c_str(), token_len,
- costs[token_index], costs[token_index], num_fuzzy_candidates, token_order, prefix_search,
+ costs[token_index], costs[token_index], max_candidates, token_order, prefix_search,
filter_ids, filter_ids_length, leaves, unique_tokens);
/*auto timeMillis = std::chrono::duration_cast<std::chrono::milliseconds>(
@@ -3168,7 +3170,7 @@ void Index::search_field(const uint8_t & field_id,
all_result_ids_len, field_num_results, group_limit, group_by_fields,
prioritize_exact_match, concurrency, query_hashes,
token_order, prefix, drop_tokens_threshold, typo_tokens_threshold,
- exhaustive_search, min_len_1typo, min_len_2typo);
+ exhaustive_search, min_len_1typo, min_len_2typo, max_candidates);
}
}