sanikolaev
commented
9 months ago Can you please provide a minimal reproducible example which includes create table, insert into and your search query. E.g. like this:
➜ ~ mysql -P9306 -h0
Welcome to the MySQL monitor. Commands end with ; or \g.
Your MySQL connection id is 23358
Server version: 6.2.13 267b05c3a@24012222 dev (columnar 2.2.5 1d1e432@231204) (secondary 2.2.5 1d1e432@231204) (knn 2.2.5 1d1e432@231204) git branch master...origin/master
Copyright (c) 2000, 2023, Oracle and/or its affiliates.
Oracle is a registered trademark of Oracle Corporation and/or its
affiliates. Other names may be trademarks of their respective
owners.
Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.
mysql> create table test ( title text, image_vector float_vector knn_type='hnsw' knn_dims='4' hnsw_similarity='l2' );
Query OK, 0 rows affected (0.00 sec)
mysql> insert into test values ( 1, 'yellow bag', (0.653448,0.192478,0.017971,0.339821) ), ( 2, 'white bag', (-0.148894,0.748278,0.091892,-0.095406) );
Query OK, 2 rows affected (0.01 sec)
mysql> select title, knn_dist() from test where knn ( image_vector, 5, (0.286569,-0.031816,0.066684,0.032926) );
+------------+------------+
| title | knn_dist() |
+------------+------------+
| yellow bag | 0.28146550 |
| white bag | 0.81527930 |
+------------+------------+
2 rows in set (0.00 sec)
--- 2 out of 2 results in 0ms ---
xansrnitu
glookka
I am getting a higher cosine _knn_dist value for a hit than my own standalone calculation. My standalone code `from numpy.linalg import norm from sentence_transformers import SentenceTransformer
embed_transformer = SentenceTransformer('all-MiniLM-L6-v2') cos_sim = lambda a,b: (a @ b.T) / (norm(a)*norm(b)) strs = ["Uttar Pradesh's capital city is Lucknow.","What is the capital city of Uttar Pradesh?"] vecs = embed_transformer.encode(strs) distance = 1 - cos_sim(vecs[0],vecs[1]) print(distance)`
gives distance = 0.12863826751708984
However, manticore knn search is returning .25727645 for the same query and target sentence -
{'aggregations': None, 'hits': {'hits': [{'_id': '2740690609156129169', '_knn_dist': 0.25727645, '_score': 1, '_source': {'content': "Uttar Pradesh's capital city is " 'Lucknow.', 'meta': {'chunk_id': 1, 'title': 'india'}}}, {'_id': '2740690609156129172',Also often time, I am getting cosine _knn_dist > 1.0 when I am not expecting a value > 1.0 I have tried with some other texts as well, but having the same observations.Am I interpreting this wrong? Or there is an issue here?
Thanks.