Draft: Real vectors for tests

tests/congruence_tests/test_common.py

@@ -14,9 +14,9 @@
 COLLECTION_NAME = "congruence_test_collection"
 
 # dense vectors sizes
-text_vector_size = 50
-image_vector_size = 100
-code_vector_size = 80
+text_vector_size = 384
+image_vector_size = 384
+code_vector_size = 384
 
 # sparse vectors sizes
 sparse_text_vector_size = 100

tests/congruence_tests/test_discovery.py

@@ -21,7 +21,11 @@
 
 
 def random_vector(dims: int) -> list[float]:
-    return np.random.random(dims).round(3).tolist()
+    _text_vectors = np.load("data/queries.npy", allow_pickle=True).astype(np.float32)
+    _text_vectors_unique = np.unique(_text_vectors, axis=0)
+    _text_vectors = _text_vectors_unique.tolist()
+    sampled_vectors = np.random.choice(len(_text_vectors), size=1, replace=False)
+    return sampled_vectors[0].tolist()
 
 
 @pytest.fixture(scope="module")

tests/congruence_tests/test_recommendation.py

@@ -24,7 +24,11 @@ class TestSimpleRecommendation:
     __test__ = False
 
     def __init__(self):
-        self.query_image = np.random.random(image_vector_size).tolist()
+        _text_vectors = np.load("data/queries.npy", allow_pickle=True).astype(np.float32)
+        _text_vectors_unique = np.unique(_text_vectors, axis=0)
+        _text_vectors = _text_vectors_unique.tolist()
+        sampled_vectors = np.random.choice(len(_text_vectors), size=1, replace=False)
+        self.query_image = sampled_vectors[0].tolist()
 
     @classmethod
     def simple_recommend_image(cls, client: QdrantBase) -> list[models.ScoredPoint]:

tests/congruence_tests/test_search.py

@@ -22,9 +22,14 @@ class TestSimpleSearcher:
     __test__ = False
 
     def __init__(self):
-        self.query_text = np.random.random(text_vector_size).tolist()
-        self.query_image = np.random.random(image_vector_size).tolist()
-        self.query_code = np.random.random(code_vector_size).tolist()
+        _text_vectors = np.load("data/queries.npy", allow_pickle=True).astype(np.float32)
+        _text_vectors_unique = np.unique(_text_vectors, axis=0)
+        _text_vectors = _text_vectors_unique.tolist()
+        sampled_vectors = np.random.choice(len(_text_vectors), size=3, replace=False)
+
+        self.query_text = _text_vectors[sampled_vectors[0]]
+        self.query_image = _text_vectors[sampled_vectors[1]]
+        self.query_code = _text_vectors[sampled_vectors[2]]
 
     def simple_search_text(self, client: QdrantBase) -> list[models.ScoredPoint]:
         return client.search(

tests/fixtures/points.py

@@ -10,16 +10,24 @@
 from qdrant_client.local.sparse import validate_sparse_vector
 from tests.fixtures.payload import one_random_payload_please
 
+_text_vectors = np.load("data/text.npy")
+_text_vectors_unique = np.unique(_text_vectors, axis=0)
+_text_vectors = _text_vectors_unique.tolist()
 
-def random_vectors(
-    vector_sizes: Union[dict[str, int], int],
-) -> models.VectorStruct:
+
+def random_vectors(vector_sizes: Union[dict[str, int], int], idx=None) -> models.VectorStruct:
     if isinstance(vector_sizes, int):
-        return np.random.random(vector_sizes).round(3).tolist()
+        if idx:
+            return _text_vectors[idx]
+        else:
+            return np.random.random(vector_sizes).round(3).tolist()
     elif isinstance(vector_sizes, dict):
         vectors = {}
         for vector_name, vector_size in vector_sizes.items():
-            vectors[vector_name] = np.random.random(vector_size).round(3).tolist()
+            if idx:
+                vectors[vector_name] = _text_vectors[idx]
+            else:
+                vectors[vector_name] = np.random.random(vector_size).round(3).tolist()
         return vectors
     else:
         raise ValueError("vector_sizes must be int or dict")
@@ -100,7 +108,9 @@ def generate_points(
     if skip_vectors and isinstance(vector_sizes, int):
         raise ValueError("skip_vectors is not supported for single vector")
 
+    sampled_vectors = np.random.choice(len(_text_vectors), size=num_points, replace=False)
     points = []
+
     for i in range(num_points):
         payload = None
         if with_payload:
@@ -115,7 +125,7 @@ def generate_points(
         elif multivector:
             vectors = random_multivectors(vector_sizes)
         else:
-            vectors = random_vectors(vector_sizes)
+            vectors = random_vectors(vector_sizes, sampled_vectors[i])
 
         if skip_vectors:
             if random.random() > 0.8: